Utilization of patient acuity data for predicting staffing needs

The LDS Hospital in Salt Lake City, Utah wishes to utilize the computerized patient acuity data to determine the staffing requirement of the nursing units. Acuity items are the nursing care procedures provided the patient during a shift; a standard time is assigned to each procedure. A model was designed which assumed the total hours of nursing care received by the patients during the previous 24 hours would be an adequate estimate of the total hours of care that would be needed in the next 24 hour period. The purpose of this study was to evaluate this current model and to develop an additional model for comparison. Two separate criteria were applied to the current model; one that looked at the overall prediction error and the other that focused on the prediction error per nurse. Both units could not meet the overall criterion a high percentage of the time but the error per nurse was significantly less on the larger unit than on the smaller unit. In addition, the following four assumptions of the current model were tested: (1) the acuity data is accurate (2) the workload distributions among the shifts are relatively stable, (3) census fluctuation will not significantly affect the model, and (4) the model serves the units equally well. Only the first assumption proved to be correct. Linear regression models were formulated to predict staffing needs on two medical/surgical units, one a 46 bed unit and the other a 32 bed unit. The units were studied for close to three months. Patient acuity data were utilized in the formulations of these models along with the census and the net change in census. The linear models were compared to the model currently being used by the hospital. The smaller unit showed an improvement in the degree of prediction errors when using the linear model; however, the larger unit did not. It was concluded that neither the current model nor the linear models explained enough of the variability in the acuity totals to adequately predict the staffing needs of the hospital.

THE UTILIZATION OF PATIENT ACUITY DATA FOR PREDICTING STAFFING NEEDS by Julia Ann Rossi A thesis submitted to the faculty of The University of Utah in partial fulfillment of the requirements for the degree of Master of Science Department of Medical Informatics The University of Utah December 1986 Copyright © Julia Ann Rossi 1986 All Rights Reserved THE UNIVERSITY OF UTAH GRADUATE SCHOOL SUPERVISORY COMMITTEE APPROVAL of a thesis submitted by Julia Ann Possi This thesis has been read by each member of the following supervisory committee and by majority vote has been found to be satisfactory. July 25, 1986 Chainnan: July 25, ]986 July 25, 1986 I l · hL~- Ii ,Ii / I}/ hi t /.1: , IV' I T. Allen Pryor D. ~imothy Bishop Jean Miller THE UNIVERSITY OF UTAH GRADUATE SCHOOL FINAL READING ·APPROVAL To the Graduate Council of The University of Utah: 1 have read the thesis of Julia Ann lbssi in its final form and have found that (I) its format. citations, and bibliographic style are consistent and acceptable; (2) its illustrative materials including figures, tables, and charts are in place: and (3) the final manuscript is satisfactory to the Supervisory Committee and is ready for submission to the Graduate SchooL D. Timothy Bishop Member. Supervisory Committee Approved for the Major Department f l '~f71..,~~ Homer R. Warner Chairman! Dean Approved for the Graduate Council ABS1RACT The LDS Hospital in Salt Lake City, Utah wishes to utilize the computerized patient acuity data to determine the staffing requirement of the nursing units. Acuity items are the nursing care procedures provided the patient during a shift; a standard time IS assigned to each procedure. A model was designed which assumed the total hours of nursing care received by the patients during the previous 24 hours would be an adequate estimate of the total hours of care that would be needed in the next 24 hour period. The purpose of this study was to evaluate this current model and to develop an additional model for comparison. Two separate criteria were applied to the current model; one that looked at the overall prediction error and the other that focused on the prediction error per nurse. Both units could not meet the overall criterion a high percentage of the time but the error per nurse was significantly less on the larger unit than on the smaller unit. In addition, the following four assumptions of the current model were tested: (1) the acuity data is accurate (2) the workload distributions among the shifts are relatively stable, (3) census fluctuation will not significantly affect the model, and (4) the model serves the units equally well. Only the first assumption proved to be correct. Linear regressIon models were formulated to predict staffing needs on two medical/surgical units, one a 46 bed unit and the other a 32 bed unit. The units were studied for close to three months. Patient acuity data were utilized in the formulations of these models along with the census and the net change in census. The linear models were compared to the model currently being used by the hospital. The smaller unit showed an improvement In the degree of prediction errors when using the linear model; however, the larger unit did not. It was concluded that neither the current model nor the linear models explained enough of the variability in the acuity totals to adequately predict the staffing needs of the hospital. v TABLE OF CONTENTS ABSTRACT ..... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 V LIST OF TABLES ...................................... Vlli LIST OF FIGURES ...................................... IX ACKNOWLEDGEMENTS X Chapter 1. INTRODUCTION 1 The Staffing Models Used at LDS Hospital ........... 2 The Research Questions Addressed ................. 9 Notes .......................................... 11 2. DESCRIPTION OF THE STUDY AND THE DATA COLLECTED Design ....................................... . Source of Data ................................. . Data Collection ............................... . . . Data Collected ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Notes 13 13 16 17 18 21 3. 1vffiTHOD OF DATA ANALYSIS ...................... 22 Evaluating the Current Model ..................... 22 Analysis of the Assumptions ..................... 23 The Development of the Linear Models ............. 25 Comparison of the Current and Linear Models . . . . . .. 26 Notes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 27 4. EV ALUATION OF THE CURRENT MODEL ... . . . . . . . . . 28 Prediction Errors of the Current Model ............ 2 8 Analysis of Four Assumptions .................... 3 1 5. DEVELOPMENT AND EVALUATION OF THE LINEAR REGRESSION MODELS ................ 4 1 The Development of the Linear Models . . . . . . . . . . . . 4 1 Testing of the Linear Models .................... 4 7 Comparison of the Linear and Current Models . . . . .. 52 6. DISCUSSION AND CONCLUSION .......... . . . . . . . . . . 56 Interpretation of Results ......................... 5 6 Limitations .................................... 62 Conclusion ..................................... 64 Recommendations .............................. 65 Notes ......................................... 67 Appendices A. DEFINITIONS OF SELECTED ACUITY ITEMS 68 B. EXAMPLE OF A UNIT ACUITY REPORT. . . . . . . . . . . . . . 73 C. THE STAFFING ASSESSMENT TOOL. . . . . . . . . . . . . . . . . 75 D. WRITTEN COMMUNICATIONS WITH THE NURSES ON WEST 5 and NORTH 6 . . . . . . . . . . . . . . . . . . . . . . . . . 79 LITERA TURE CITED ................................... 82 V11 LIST OF TABLES Table Page 4.1. Statistics on the First and Second Criterion Applied to the Current Model Prediction Errors on West 5 and North 6 .......................... 29 4.2. Comparison of the Actual Number of Nurses on West 5 and the Number of Nurses Claimed by the Total Acuity Score ................... . . . . . . . . . . .. 32 4.3. Comparison of the Actual Number of Nurses on North 6 and the Number of Nurses Claimed by the Total Acuity Score ............................... 3 3 4.4. Comparison of the Net Change in Census Between West 5 and North 6 ............................. 36 5.1. Correlation Coefficients of the Independent Variables Regressed on the West 5 Evening Shift Acuity Score .. 42 5.2. Correlation Coefficients of the Independent Variables Regressed on the North 6 Evening Shift Acuity Score .. 43 5.3. The Correlation Matrix of the West 5 Variables ................................. 45 5.4. The Correlation Matrix of the North 6 Variables ................................ 46 5.5. Analysis of the Linear Regression Variance on the West 5 Original Data Set .................... 48 5.6. Analysis of the Linear Regression Variance on the North 6 Original Data Set .... . . . . . . . . . . . . . .. 48 LIST OF FIGURES Figure 4.1. Distributions of the 24 hour acuity workload . . . . . . . . . . . . . . . . . . . . . . . . . 3 5 4.2. Correlation between the net change in census and the prediction errors on North 6 .............................. 3 7 4.3. The relationship between the previous 24 hour acuities and the evening shift acuities on West 5 ............................. 3 9 4.4. The relationship between the previous 24 hour acuities and the evening shift ac uities on North 6 ............................ 3 9 5.1. The prediction errors of the linear regression model on the West 5 original data set ....... 49 5.2. The prediction errors of the linear regression model on the North 6 original data set ........ 49 5.3. The prediction errors of the linear regression model on the West 5 test set data ............... 5 1 5.4. The prediction errors of the linear regression model on the North 6 test set data ......... . . . . . . 51 5.5. The prediction errors of the linear model and the current model on the West 5 test set data .................................. 5 3 5.6. The prediction errors of the linear model and the current model on the North 6 test set data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 5 3 ACKNOWLEDGMENTS My deepest appreciation goes to Dr. D. Timothy Bishop for his excellent guidance and continual support throughout this project. In addition, I wish to thank Dr. Simon Tavare, Martha Veranth, 'Pamela Shelton and my advisor, Dr. T. Allen Pryor, for their generous assistence. Finally, a special thanks to my devoted husband, Dr. Hugo Rossi, and son, Raffaele, for their constant encouragement and unending patience. CHAPTER 1 INTRODUCTION In 1965, Harvey Wolfe and John P. Young made the following statement: "Perhaps the most difficult task facing the modern hospital administrator is the effective allocation of resources and serVIces to meet rapid daily changes in the demand for care." 1 Not only is this judgment still applicable in 1986, but the most difficult task has become even more difficult due to the intensive pressure for cost containment and the emergence of prospective reimbursement by third party payers. In 1972 a conference was conducted by the Division of Nursing under the U.S. Department of Health, Education, and Welfare entitled Research on Nurse Staffing in Hospitals. Mr. John R. Griffith, professor and director of the Program and Bureau of Hospital Administration at the University of Michigan, reported on the impact of administrative and cost factors on nurse staffing. He found, to his surprise, that both the federal and state governments were relatively unconcerned about the rising medical costs and were unwilling to take real steps towards controlling those costs.2 In addition, he marveled at the total lack of insight on the part of the nursing supervisors to the problem of nurse staffing and its economic impact.3 The following quote illustrates Mr. Griffith's excellent foresight: The conclusion seems to be that although the impact of administrative and cost factors on nurse staffing is at the moment relatively insignificant, there is quite a powerful chain reaction that might occur if there is a basic change in the attitude of the American people toward the cost of hospital care. Given attention to this it is inevitable that there will be attention to nurse staffing, if for no other reason than that it is the largest single component of hospital costs. The resulting attention will lead to calls for improved technology, better defense of present practices, and evaluation of proposals for improved organizational parameters.4 2 In 1982, Grimaldi and Micheletti stated that, "Total nursing costs­essentially the salaries and fringe benefits paid to registered nurses, licensed practical nurses, and nurses aides-comprise roughly 30 percent of the typical hospital's budget for direct patient care."5 It is apparent that the chain reaction predicted. by Mr. Griffith has certainly occurred. Today, in the mid-eighties, hospitals, faced with new and stricter reimbursement methods, are turning to computers in hopes of solving the difficult problem of staffing efficiency. Administrators are seeking computer systems that can (1) calculate the amount of nursing care given so that patients can be charged individually for their care (2) schedule the nurses on a month to month basis and, (3) determine the staffing needs of each unit in order to optimize the utilization of personnel. 6 The Staffing Models Used at LDS Hospital The focus of this study concerns a staffing model recently instituted at LDS Hospital in Salt Lake City, Utah, a 550-bed tertiary care center. LDS Hospital is one of several teaching centers for the 3 University of Utah School of Medicine. This hospital has been a participant in the ongoing development of a comprehensive computer system called HELP, Health Evaluation through Logical Processing.7 The department of Medical Informatics within the University of Utah School of Medicine has been. responsible for this development; its purpose is to meet the clinical, administrative, teaching and research needs within a single computer system. The computerization of the nursing needs of the hospital is one of the many areas being developed. The goal is to have bedside terminals throughout the hospital that will allow for computerized charting, computerized nursing care plans, automatic calculation of the hours of nursing care a patient has received, and automatic billing for nursing services. In addition, the nursing administration wants to be able to use the computer for scheduling nurses and for determining the appropriate numbers of staff to be allocated to the various units. Before continuing, it is necessary to describe the organizational structure of LDS Hospital as it relates to staffing. Every nursing unit has a certain number of regular staff that routinely work on the unit. In addition, there is a group of nurses not assigned to any particular unit. They belong to the PRN (as needed) pool and they are often called the float nurses in reference to the fact that they float from unit to unit according to daily need. The decisions about the allocation of nursing personnel are made in a central staffing office that is open 24 hours a day; this office also heads the PRN pool. Nurses are asked to stay home without payor to take vacation time when there is a lower than normal census. 4 History of the Current Model A new model for determining staffing needs was introduced in January of 1986. Previously, the method used to allocate staff was based on a single study done in the mid-seventies that determined the average workload per patient for each shift. There were no means for updating the calculations of workload per patient and there were no adjustments made for the differing levels of personnel staffing the shift. The staffing office had a chart that indicated the number of staff allocated to each unit based on the census at the beginning of the shift. LDS Hospital personnel were motivated to adopt a new model in an effort to improve their staffing patterns. It is quite surprising that in the 1970s a hospital would adopt such a system considering that as early as 1960, Connor introduced the concept of patient classification systems for the purpose of identifying and quantifying nursing care requirements of patients.8 In 1965, Wolfe and Young reported that staffing by census was totally inadequate because of the extreme variation in the workload from patient to patient and even for the same patient from day to day.9 By the early 1970s, the majority of staffing models were based on classification systems and in 1973, Aydelotte summarized 200 methodologies and cited over 1,000 staffing studies. 1 0 The new model being used at the LDS Hospital, which will be referred to as the current model throughout this study, is not a classification type of model. However, there are some features of the general classification models that are in common with the development of this current model and will be discussed. 5 A classification model groups patients according to their nursing care needs. The numbers of patients in each group determine the levels and numbers of nursing staff necessary to deliver the care. Two common methods of classification systems are those that are based on prototype evaluations and those that are based on factor evaluation. I I A prototype evaluation classifies the patients by identifying typical characteristics of patients in a particular category; for example, the patient who is bedridden versus the patient who can ambulate but requires assistance versus the patient who can ambulate without assistance. In contrast, the factor evaluation identifies each nursing care item required by the patient, such as a complete bed bath or a blood transfusion; the frequency and the complexity of the care items determine the patient's classification. Once patients are classified, the approximate nursing time can be determined in two ways. The first approach involves classifying a test set of patients, calculating the average hours of care the patients in each classification receive and, finally, applying those average times to all the future patients according to their classification. I2 The second approach is to calculate the average amount of time each nursing procedure requires. This usually involves time and motion studies. The average times are multiplied by the anticipated frequency of the care item. Generally some kind of constant time is included to account for indirect care activities. 6 Finally, the times are summed up to represent the total time a patient will require. This method is often considered more objective. 13 The problem is that marking a long list of anticipated care procedures can be time-consuming and tedious. The difficulties with both methods are in deciding the number of classification groups that are appropriate, verifying that the method of determining the times is valid and reliable, and up-dating the model with changing nursing procedures and hospital policies. Description of the Current Staffin~ Model As stated earlier, the current model used by LDS Hospital is not a classification tool and it does not follow any established methodology for determining staffing needs. However, it is based on factor evahiation and a summation of the total hours of nursing care given to the patients. Therefore, some of the findings in the classification studies are relevant to this study and contribute to the final evaluation of the model used by the LDS Hospital. As the model is described, reference will be made to pertinent findings In the literature. A substantial number of nursing care procedures have been individually assigned standard times based on time and motion studies. These procedures are referred to as patient acuity items. The total hours of nursing care given to a patient during a shift are measured by summing the standard times allocated to the various items marked for the patient. This is called the patient's acuity score. In addition, all the acuity scores of the patients on a particular shift are summed to equal the unit's acuity score for that 7 shift. The three shifts can then be totaled to give the 24 hour acuity score. Appendix A lists some of the acuity items included in the acuity tool. Aydelotte identified two of the assumptions necessary for this type of approach: (1) "that nursing care consists of a serIes of procedures that are discontinuous and discrete in nature"; and (2) "procedures are performed for patients in a time sequence, and it IS possible to rearrange this sequence to make better utilization of personnel." 14 In the future, when the hospital has converted to computerized charting, these acuity items will be automatically extracted from the nurse's charting. Currently, at the end of a shift, the nurse who has cared for the patient marks the appropriate care items on a computer card and every 24 hours the cards are read into the computer. Each patient's record includes the individual acuity items marked on each shift and the total hours of care received during each shift throughout his or her hospitalization. As early as 1977 it was noted that this type of documentation would allow for nursing care to be itemized as a separate part of the patient's bill.15 More recently, a large proportion of the literature on staffing addresses the benefits of charging separately for nursing service. 16 , 17,18 A criticism of this type of approach has been the time wasted in marking long lists of nursing procedures. Since the plan is to extract this information directly from the computer charting, this can be avoided in the future. In January 1986, the nursing administration began to use the acuity totals to determine the staffing needs. An assumption was made that the total hours of care given over the past 24 hours 8 would closely reflect the hours of care needed In the next 24 hours. A description is as follows. The 24 hour acuity distribution among the three shifts was averaged from the acuity data acquired over the previous year. On an average 40% of the 24 hour acuity total was on day shift, 40% on evening shift, and 20% on night shift. There are other models, such as SCALE, that also allocate a set workload distribution for the 24 hour period; however, SCALE is set at 45% on days, 37% on evenings and 18 % on nights. 1 9 Once a day, at the end of the evening shift, the percentages of the previous 24 hour acuity total are calculated. These values are then used as estimates of the acuity scores for the next three shifts. Each nurse is assumed to work 7.6 hours. The projected acuity score for each shift is divided by 7.6 to equal the number of nurses needed. Interestingly, a review of the literature revealed that very little attention was paid to the professional level or experience of the nurse performing the task. The input from the nurses participating in this study suggests that ignoring the experience and professional level of the nurse does contribute to staffing problems. This model is unlike the classification models in that it does not involve any preassessment of the patient's needs. It assumes that the care already provided will be close to the care that will be needed. Similarly to many classification models, it does not account for the level or mix of personnel. The model makes no adjustments for the changes in census that have occurred since the total acuity scores were summed. It became apparent during this study that adjustments were being made by the staffing office. 9 A great deal of effort had gone into designing and implementing the acuity system for the purpose of documenting nursing care and charging for that care. In addition, it was recognized that the patient acuity data would be an excellent resource for research. The nursing staff had been 'promised' by the nursing administration that their efforts in filling out the rather cumbersome acuity forms, would be rewarded with an improved staffing model. The current model was chosen more as a temporary measure than as a serious solution to the staffing problems at LDS Hospital. However, if the model proved to be adequate, it would be much easier than a classification model. The Research Questions Addressed The questions addressed in this study were the following: (1) does the current model adequately predict the staffing needs of the nursing unit? (2) if not, where does it fail and (3) would a linear model using selected acuity data improve the predictions? Chapter 2 outlines the study design and the data collected. Two medical/surgical units were studied for close to three months. Chapter 3 discusses the methods used to evaluate the current model and to analyse four assumptions of the model that were potential weak' points: (1) the acuity times accurately reflect the nursing care time; (2) the workload distribution doesn't vary enough to create large errors in the predictions; (3) the census fluctuation will not be significant; and (4) the model will serve the units equally well. The results of the evaluation of the current model are presented in Chapter 4. 10 Chapter 3 also describes the methods used to develop the linear models. The inconvenience of classifying all the patients in contrast to the easy availability of the acuity data once the system is automated is the reason that the acuity data was tested in a linear model. In addition, the linear model makes essentially the same assumptions that a classification model makes: the nursing care items are additive and separable. Even though an ideal linear model assumes the independence of the variables, it was recognized that some of the acuity items would not be completely independent. This was actually verified when the correlation matrix of the variables was calculated. The decisions to select certain combinations of variables for the linear regression models implies that only those having low correlation coefficients would be selected. A criticism of classification systems has been that there is overlap in patient classes because the nurSIng care procedures are not independent.20 The linear models chosen, their performance on test set data, and their statistical comparison with the current model are presented in Chapter 5. Finally, Chapter 6 interprets the findings and discusses the limitations of this study. Recommendations are offered for future study and certain features are identified that should be included in a staffing model. 1 1 Notes 1 Harvey Wolfe and John P. Young, "Staffing the Nursing Unit Part 1. Controlled Variable Staffing," Nursini Research 14, no.3 (Summer 1965): p. 236. 2 John R. Griffith, Impact of Administrative and Cost Factors on Nurse Staffini, U.S., DHEW Publication, no.(NIH) 73-434. (Report of the May 1972 Conference: Research on Nurse Staffing in Hospitals), p. 94. 3 Griffith, p. 92. 4 Griffith, p. 95. 5 Paul L. Grimaldi and Julie A. Micheletti, "RIMs & the Cost of Nursing Care," Nursini Manaiement 13, no. 12 (December 1982): p. 12. 6 Marylou Kiley, Edward Halloran, Jerry Weston, Judy Ozbolt, Harriet Werley, Marjorie Gordon, Phyllis Giovanetti, John Thompson, Roy Simpson, Rita Zielstorff, Joyce Fitzpatrick, H. Stephanie Davis, Margot Cook, and Margaret Grier, "Computerized Nursing Information Systems (NIS)," Nursini Manaiement 14, (July 1983): p. 28. 7 T. Allen. Pryor, Reed M. Gardner, Paul D. Clayton, and Homer R. Warner, "The HELP System," Journal of Medical Systems, 7, no.2 (1983): p. 83. 8 R.J. Connor, "A Hospital Inpatient Classification System" (Ph.D. diss., The Johns Hopkins Univ.,1960). 9 Wolfe and Young, p. 237. 10 Myrtle K. Aydelotte, Nurse Staffini Methodology: A Review and CritiQ.ue of Selected Literature, U.S., DHEW no.(NIH)73- 433. 11 Faye G. Abdellah and Eugene Levine, Better Patient Care Through Nursing Research (New York: The Macmillan Company, 1965). 12 12 Phyllis Giovannetti, "Understanding Patient Classification Systems, " Journal of Nursing Administration, (February 1979): p. 5. 13 Giovannetti, p. 6. 14 Aydelotte, p. 52. 15 Margaret Williams, "Quantification of Direct Nursing Care Activities." Journal of Nursing Administration 7, (October 1977) 15-18. 16 Grimaldi and Micheletti, p. 12. 17 Joan Trofino, "A Reality Based System for Pricing Nursing Service." Nursing Management 19, (January 1986): p. 19. 18 Hollie Vanderzee and George Glusko, "DRGs, Variable Pricing, and Budgeting for Nursing Services," Journal of Nursing Adn1inistration (May 1984): p. 11. 19 Aydelotte, p. 49. 20 Robert Vaughan and Vernon MacLeod, UN urse Staffing Studies: No Need to Reinvent the Wheel," Journal of Nursing Administration, (March 1980): p. 10. CHAPTER 2 DESCRIPTION OF THE STUDY AND THE DATA COLLECIED As stated in Chapter 1, this study addresses the following three questions: (1) how well does the current hospital system of prediction fit the staffing needs of the nursing units, (2) what are the circumstances in which the current system fails, and (3) does a linear regression model significantly improve the predictions? Design Data were collected from two nursing units at LDS Hospital for a period greater than two months. Data included the total acuity, the census, the change in census, the number of staff, and a tally of particular acuity items. In addition, at the end of every shift, a staffing evaluation questionnaire was filled out by the nurse in charge in an effort to identify the shifts where the acuity totals were unlikely to accurately reflect the total hours of care needed. The intent was to be able to adjust the acuity total to better evaluate the degree of prediction error, where prediction error represents the difference between the actual acuity total and the predicted acuity total. Unfortunately, the response rate to the assessment tool was not sufficient. However, a comparison was made between the assessments and the prediction errors to determine the level of agreement. The collected data were utilized for the following purposes: 14 1. Evaluation of the degree of prediction errors made by the current model. These errors were tested against two separate criteria described below. The percentage of times each criterion was not met was calculated. 2. Analysis of the underlying assumptions of the current model. The specific assumptions tested were: (a) the acuities are accurate, (b) the workload distributions are reasonably stable and similar on all the units, (c) the shift to shift census fluctuations will not be significant enough to adversely affect the predictions, and (d) a single prediction model will serve all the units equally well. 3. Development of a linear model to predict the total acuity. To accomplish this, independent variables were examined to determine their individual and combined predictive value. The variables were tested separately on each nursing unit, and a correlation matrix of the variables was calculated for each unit. A single linear regression model was selected for each unit. 4. Statistical comparison of the linear regression model with the current model. The linear regression model was. generated by the data collected during the early part of the study (this is referred to as the original data set). The remainder of the data were available for testing the linear models. In addition, the test set data were used to compare the linear model to the current model. Statistical tests were applied to test for significant differences between the two models on both units. 15 In order to measure the adequacy of the current model's prediction, two distinct criteria were used. The first criterion was the absolute value of the difference between the acuity prediction and the actual acuity score; here eight hours was considered the maximum admissable difference. This was chosen because it would, like the current model, not distinguish between the units. Since the maximum allowed deviation is the equivalent of one nurse's shift, this criterion reflects the hospital's assessment of an allowable variation of cost per unit per shift. The second criterion was the ratio of this difference to the numbers of nurses present, l.e., the variation between predicted and actual nursing care per nurse. This criterion will distinguish the large from the small unit and allows for examination of the different impact of the same prediction errors on the two units. Furthermore, it is based on the nurses' assessment of allowable variation in the work load per nurse. While formulating the linear models, it was decided to focus on predicting the acuity for one particular shift, rather than looking at all three shifts. The total acuity score of the evening shift was selected as the dependent variable to be predicted. This shift was selected for the following two reasons. Prior to using the current model, the evening shift was thought to be less busy than the day shift and had always received less staffing. Also, the evening shift was 16 hours away from the end of the 24 period used to predict the acuities, and hence could be expected to be the most deviant from the predictions. In addition, only the evenIng shift was used to compare the linear model with the current model. Source of Data Two nursIng units, West 5 and North 6, were selected for data collection. These particular units were chosen for the following reasons: 1. supportive attitude of the head nurses; 2. agreement among the staff to participate; 3. dissimilarity between the units in the types of patients served, the available bed space, and the fluctuation In census; and 4. a history of consistent and relatively accurate usage of the acuity tool. 16 West 5 IS a forty-six bed medical/surgical unit described as the acute care unit of the cardiovascular center. The patient population includes persons admitted preoperatively for open heart surgery, persons transferred out of the thoracic and coronary intensive care units, and patients that require telemetry for any reason. Typically 20% to 30% of the patients on the unit are being monitored. The unit occasionally takes over-flow patients from other units. West 5 had a 93% occupancy rate from January of 1986 until the end of March. The average length of stay for a patient is five days but the range of stay is less than one day to more than three months. The number of full time equivalent staff members is 38; 88% are RNs with the remaining 12% being LPNs. The average nurse per patient ratio on day shift is one nurse to four patients, on evening shift it is one nurse to five patients and on night shift it is one nurse to seven patients. The nursing management style is primary care nursing. Each shift does have a nurse in charge who takes a lighter patient assignment and keeps an overview of the unit. North 6 is a thirty-two bed general surgery unit. The 17 occupancy rate is 69% and the average length of stay is three to four days. Only rarely do they have a patient that would stay longer than ten days. The number of FTE's is eighteen. RNs make up 84% of the staff with the remaining 16% being LPNs. The average nurse per patient ratio on day and afternoon shifts is one nurse to four patients while on night shift it's one nurse to nine patients. The patient population includes persons admitted for gastric-bypass, gallbladder, and bowel surgery. Occasionally, the unit takes over­flow patients, in particular, patients admitted for cardiac catherization. The nursing management style is the same as for West 5. Data Collection Separate meetings were held with the head nurse and assistant head nurse of each unit to describe the study and to request permission to collect data from their units. Initial agreement to participate was dependent upon the consensus of the nursing staff members. The study was outlined in a presentation given at the next scheduled staff meeting. Upon agreement, a starting date was scheduled. Copies of a written communication were distributed among the staff which reviewed the study outline, the assessment tool, and the location of the envelope provided for collection of the tool. This communication also introduced the study plan to those staff members unable to attend the staff meeting. Samples of written communications are listed in Appendix D. 18 During the first week, the nurse in charge of the unit was called at the end of each shift and reminded to fill out the form. A telephone number was listed on the collection envelope for questions or concerns. During the three months of data collection, several notices and reminders were distributed to address questions that had arisen and to encourage participation. The data on West 5 was collected from January 7, 1986 until March 21, 1986, 74 samples of each shift for a total of 222 shifts. The data on North 6 was collected from February 12, 1986 until April 21, 1986, 69 samples of each shift for a total of 207 shifts. Data Collected The following data items were collected: Unit Acuity Report A unit acuity report is available for each unit on each shift and lists the patienfs names, their acuity score, the total acuity score for the unit and the average acuity score for the shift. The report also includes a special marker if the patient has been transferred, admitted, discharged, or off the unit for some part of the shift. On the evening shift acuity report, the 24 hour acuity for the entire unit and the average 24 hour acuity per patient is listed. Appendix B contains an example of an evening acuity report. 19 Census and the Change in Census The census data represents the census at the very beginning of the day shift and does not include anticipated admissions, transfers, or discharges. Change in census refers to the change in census from the beginning of one shift to the beginning of the next shift. The census data was copied from the nursing unit records and periodically verified with the staffing office records. A Tally of Individual Acuity Items A computer program was written that searched the patients' records for the individual acuity items on a particular shift. These acuity items were then tallied to give the total number of times each task was performed during the shift. Appendix A lists and defines each acuity item extracted. Staffing Assessment Tool At the end of every shift, the nurse in charge was asked to fill out the staffing assessment tool listed in Appendix C. This tool is a modified version of the Unit Staffing/Care Evaluation Questionnaire written by Williams and Murphy. 1 Their study focused on the staff nurses' perceptions of staffing adequacies and was part of a larger study done at San Joaquin General Hospital in 1976 to develop methods for improving the use and effectiveness of nursing personnel. For the purpose of this study, the information obtained from the staffing assessment tool was to identify those shifts where the 20 actual acuities were inadequate. The initial intent was to make adjustments in the acuity scores; however, because of the low response rate, no alterations were made. The information proved to be helpful in the final analysis of the study. Numbers and Levels of Nurses A record book was kept on the units which listed the numbers and levels of the nurses staffing the unit. The patient assignment sheet was periodically consulted to verify this information. Notes 1 Laurel N. Murphy, Marjorie S. Dunlap, Margaret A. Williams, and Marylou McAthie, "Methods for Studying Nurse Staffing in a Patient Unit," U.S., DREW no. HRA 78-3 (May 1978) p. 47. 21 CHAPTER 3 METHOD OF DATA ANALYSIS The previous chapter described the study itself and the data collected. This chapter focuses on the methods used to (1) evaluate the extent of the prediction errors in the current model, (2) analyse the current model's assumptions, (3) derive a linear model, and (4) compare the linear model with the current model. Evaluating the Current Method A prediction error refers to the difference between the actual acuity and the predicted acuity. Throughout this study the predicted acuities were subtracted from the actual acuities so that the negative prediction errors represent overpredictions. The adequacy of the current model was evaluated separately USIng the following two criteria: (1) that the absolute value of the prediction error be less than or equal to eight hours, and (2) that the absolute value of the ratio of the prediction error to the number of nurses be less or equal to one hour (see Chapter 2). The original databases were used for testing the current model. The means and standard deviations of the prediction errors were calculated and compared from shift to shift and unit to unit. The percentage of times that the prediction errors were outside the set criterion was calculated for each unit and each shift. A conlparision was made between the results of the two criterion tested. Analysis of the Assumptions The Accuracy of the Acuity Scores 23 The current method of predicting staffing at LDS Hospital assumes that the acuity scores are accurate. It was not the purpose of this study to validate the acuity tool. However, the ultimate goal in predicting the total acuity is to be able to predict the number of nurses needed. Therefore, a strong positive linear relationship is expected between the hours of care claimed and the number of nurses actually available and, in addition, the hours of nursing care claimed should be reasonably close to the number of staff that were available. The following method was used to analyse the relationship of the total acuity score and the actual number of nurses staffing the unit. The total acuity scores were extracted from shifts staffed with nIne to thirteen nurses on West 5 and shifts staffed with three to seven nurses on North 6. Since the current method assumes that each nurse staffing the unit works approximately 7.6 hours, the total acuities were divided by this number to equal the approximate number of nurses the acuity score represented. The mean number of nurses the acuity totals claimed and the standard deviations of that value were compared to the actual number of nurses. At-test was performed on the differences between the claimed and the available hours of nursing care to look for deviation other than randomness. The claimed hours of nursing care turned out to be reasonable with relatively low standard deviations (see Results, Chapter 4). Consistency of the Workload Distributions Another assumption of the current method is that all the 24 units will have a 24 hour workload distribution of 40% on day shift, 40% on evening shift, and 20% on night shift. Variability in the distributions would have an important effect on the staffing patterns. Histograms of each shift's percentage of the 24 hour acuity total were studied for the two units in order to visualize their workload distributions. The results showed that one of the units had the assumed distribution while the other unit differed on the day and evening shifts. Both units showed considerable variation within each shift. The Effect of Census Fluctuations The census change between shifts is not accounted for in the current method (Chapter 1). Because every patient represents at least half an hour of nursing care and usually one to two hours, a change in the census will have an effect on the total acuity scores. The degree of census change was evaluated as well as its correlation with the prediction errors. A comparison was done on the net change in census between day shift and afternoon shift for each unit. North 6 had considerably larger census fluctuations than did West 5. Therefore, shifts with considerable census changes, and/or large prediction errors were extracted from the North 6 database 25 and were correlated with each other. A reasonably good correlation was found but there were also points with large prediction errors and little or no change in census. The Generality of the Current Model An important assumption made by the current method is that the hospital only needs one predictive model for all the nursing units. If this is a valid assumption, then the predictive value of the previous 24 hour acuity should be fairly similar on both units. This was first tested by treating 40% of the previous 24 hour acuities as independent variables in linear regression models. Secondly, 40% of the previous acuities on both units were plotted against the evening acuities to visually compare the relationships. The Development of the Linear Models A statistical package called LIDA was utilized for data analysis and linear regressions. 1 Linear regressions were performed on each individual variable against the evening shift acuity total. LIDA lists the correlation coefficient and the tail probability for the exclusion of the independent variable. In addition, the predicted acuities are listed and the residual values (the differences between the actual acuities and the predicted acuities). Those individual variables with a p-value larger than 0.1 were discarded. A correlation matrix of the remaining variables was examined to determine the level of dependence of a given variable relative to any other variable. Variables with low correlation 26 coefficients were then combined and the linear regressIons were repeated. The regressions were repeated in this manner, increasing the number of variables up to four. The rules of low correlation and high r2 values were not strictly followed. In several instances, variables with high correlation worked well in combination. For example, on North 6 the census was highly correlated with the day shift acuity; however they worked well together in the linear regression model. The linear combinations that produced the highest r2 values were then utilized to predict staffing on a test set of data (a separate set of data not used in the calculation of the linear equation). The prediction errors were plotted against the actual acuities. To assess how data sensitive the equations were, the regressions were reestimated using the original data combined with the test set data. The new linear equations were examined for variations from the original equations. Comparison of the Current and Linear Models Finally, the linear model was compared to the hospital model for predicting the acuities on the test set data. The prediction errors of both methods were plotted together. Statistical tests were performed to check for significant differences in the two methods. The methods were compared both on their tendencies to over- or underpredict and on the ranges and standard deviations of their prediction errors. 27 Notes IJohn A. Hartigan, LIDA, computer software (Yale University Department of Statistics, 1983). Disk. CHAPTER 4 EV ALUATION OF THE CURRENT MODEL One of the three objectives in this study was to determine the adequacy of the current prediction model that is used by LDS Hospital to project staffing needs. The differences between the predicted and the actual acuities were evaluated and four assumptions of the current model were tested. Chapter 3 described the methods applied and this chapter presents the results of those methods. Prediction Errors of the Current Model As explained in Chapter 3, a prediction error is the difference between the actual acuity and the predicted acuity. Negative values represented overpredictions. The original data set was used to test the current model. The prediction errors were analysed twice, using two different criteria: the criterion of plus or minus eight hours for the total acuity score, and the criterion of not more than plus or minus one hour of nursing care per nurse. The first criterion of a predicted acuity within eight hours of the actual acuity was tested on both units. The mean and standard deviations of the prediction errors for the two units and the percentage of times that were outside the criterion are listed in the first three rows of Table 4.1. Table 4.1 Statisitics on the First and Second Criterion Applied to the Current Model Prediction Errors on West 5 and North 6 29 Night Shift Day Shift Evening Shift Mean Pred. Error for the Shift Acuity Std. of the Shift Pred. Error Mean Pred. Error per Nurse Std. of the Pred. Error per Nurse % of Errors Not Meeting 1 st Criteriona % of Errors Not Meeting 2nd Cri terion b W5 N6 2.2 0.5 4.3 3.3 0.3 0.2 0.7 1.2 15% 3% 15% 32% W5 N6 W5 N6 4.4 -2.6 -6.1 0.9 6.7 11.2 8.5 9.3 0.3 -0.3 -0.6 0.3 0.5 1.8 0.7 1. 7 26% 40% 56% 43% 11 % 46% 15% 59% a The 1st criterion equals plus or minus 8 hrs for the shift's acuity total. b The 2nd criterion equals a prediction error that does not represent more than plus or minus 1 hr per nurse. 30 The means of the prediction errors indicated that the West 5 day shift tended to be underpredicted while the evening shift tended to be overpredicted. This was consistent with the West 5 workload distribution which was higher on the day shift and lower on the evening shift than the model assumed. North 6 had the higher standard deviations of the prediction errors. The night shifts had the lower percentage of prediction errors outside the eight hour criterion and the evening shifts had the highest percentage. The prediction errors were reevaluated using the second criterion. Each prediction error was divided by the number of staff that the current model allowed. Line 6 of Table 4.1 lists the percentage of times the absolute value of the prediction error per nurse was greater than one hour. Lines three and four give the means of the prediction errors per nurse and the standard deviation. The percentage of times the units were outside of the second criterion differed strikingly from the first criterion. Where the West 5 day shift failed to meet the first criterion 26% of the time, it failed only 11 % of the time on the second. The North 6 evening shift, on the other hand, did not meet the first criterion 43% of the time and could not meet the second criterion 59% of the time. The North 6 percentages are much higher than the West 5 percentages on all three shifts. The second criterion improved the percentages for West 5 and worsened the percentages for North 6. This is not surprising since North 6 is the smaller unit and will have fewer staff. The same prediction error for both units therefore represents a larger burden per nurse on the smaller unit. 31 The mean of the prediction error per nurse was small for both units on all three shifts; but North 6's standard deviations are very large compared to those of West 5. This helps to explain the difficulty which North 6 experienced with the current model. If the prediction errors, even occasionally, represent almost two hours of work per nurse, then the reason for dissatisfaction is clear. These wide standard deviations of the prediction errors suggest a lack of flexibility in the model that contributes to its inability to adjust to the normal variations inherent within the hospital environment. The analysis of the assumptions in the following section explore this inflexibility. Analysis of Four Assumptions Because of the current model's poor performance, it was decided to test four significant assumptions of the model. This section describes the results of those tests. Accuracy of the Acuity Scores The shifts staffed with nine to thirteen nurses were extracted from the West 5 database and those staffed with three to seven nurses were extracted from the North 6 database. The acuity scores for each shift were converted from hours of care to numbers of nurses by applying an assumption made by the current model: every nurse works approximately 7.6 hours. Tables 4.2 and 4.3 compare the available nurses and the number of nurses the acuity scores would have called for. The t-values were calculated and the differences between claimed and available were tested for 32 significance at the 5% level. Initially it appeared that the tendency in almost every instance was to overclaim. However, when the units had the higher numbers of staff, there was not a significant difference. This suggests that, in order to get the work done that needs to be done, the nurses actually worked more than the 7.6 hours or that they did their work faster than the acuity times allowed. Table 4.2 Comparison of the Actual Number of Nurses on West 5 and the -Number of Nurses Claimed by the Total Acuity Score Actual number of nurses 9 10 1 1 12 Mean number acuity calls for 10.2 10.5 11.3 11.9 Std of the number acuity calls for 1.6 1.2 0.9 1.2 Significance at the 5% level yes yes yes no Sample Size 10 18 34 43 *West 5 data from Jan. 7 through Feb. 11. 13 13.1 0.9 no 49 Table 4.3 Comparison of the' Actual Number of Nurses on North 6 and the Number of Nurses Claimed by the Total Acuity Score Actual number of nurses 3 4 5 6 Mean number acuity calls for 3.1 4.4 5.3 6.6 Std of the number acuity calls for 0.55 0.85 0.77 0.84 t-value 1.68 2.05 2.13 3.84 Significance at the 5 % level yes yes yes yes Sample Size 51 20 40 28 *N orth 6 data from Feb. 12 through Apr 31. Stability of the Workload Distributions 33 7 7.1 0.54 0.73 no 8 As stated in Chapter 4, the hospitals' current method for predicting staffing assumes that the distribution of the 24 hour total acuity is 40% on day shift, 40% on evening shift, and 40% on night shift. The model multiplies the previous 24 hour total acuity by these assumed percentages once a day, at the end of the evening shift. The resulting values are the predicted acuities of the three subsequent shifts. Therefore, even if the previous 24 hour acuity IS similar to the next 24 hour acuity, but the workload distributions are not the same, then the model will be inaccurate. 34 Figure 4.1 shows the histograms of the actual workload distribution for West 5 and North 6, respectively. The graphs represent the entire data collected, 222 shifts for West 5 and 207 for North 6. It can be seen that the average workload distribution for West 5 is closer to 45% on day shift and 35% on evening shift. The following example illustrates the prediction error this percentage represents. If a 24 hour acuity were 240 hours, the 5% error would equal a 12 hour shortage on day shift and a 12 hour overprediction on evening shift. This is equivalent to 1.6 nurses. The current method could potentially allow for adjusting the basic distributions from unit to unit but it would not allow much day to day variation in the distributions. Therefore, the ranges and widths of the distributions are more important than the means. In general, the ranges for West 5 are not as wide as the 6 Norths ranges. The majority of the North 6 night shifts account for 15% to 25% of the workload, but the range extends from 5% to 55%. Interestingly, the North 6 evening shifts have the same range as the night shifts despite the large differences in their mean percentage of the workload. The current model's inability to adjust to these variations is a major weakness of the model. Even if the bulk of the distribution on evening shift is between 35% and 40%, the significant number of shifts that are outside that range would cause considerable error using the current prediction method. Suppose the 24 hour acuity total were 160 hours, then prediction for the following day and evening shift would be 64 hours. Even if the following 24 hours had the same total acuity of 160 hours, if the day shift did 50% of the workload and the evening only 30% of the 35 workload, then the day shift would require 10.5 nurses and the evening would require 6.5 nurses. However, the model would have given each shift 8.5 nurses. It is likely that extremes in the workload distributions are caused by census fluctuations; a factor which is not accounted for in the current model. 60 50 -a- Night Count 40 -+- Day 30 -0- Evening 20 10 0 0 10 20 30 40 50 60 70 Percentage of the Workload on West 5 40 30 Count 20 -a- Nights -+- Days -0- Evenings 10 o 10 20 30 40 50 60 70 Percentage of the Workload on North 6 Figure 4.1 Distributions of the 24 hour acuity workload. 36 The Importance of Including Census Fluctuations Table 4.4 gives a statistical summary of the net change in census between the day shift and the afternoon shift for both units. This data was obtained from the original data set. It should be noted that the net change in census did not reflect the overall number of admissions and discharges on a shift. For example, it could be that five patients were admitted while six were discharged on the same shift, showing only a census change of one. North 6 had a wider range of census change and a larger standard deviation indicating an overall larger census fluctuation than West 5. Statistics Minimum Maximum Mean Std. Table 4.4 Comparison of the Net Change in Census Between West 5 and North 6 West 5 ( 46 bed unit) North 6 (32 -6 -10 6 5 0.14 -1.7 2.3 3.8 *Data from the original data sets. bed unit) 37 The thirty-five evenIng shifts that had prediction errors greater than or equal to plus or minus ten hours and shifts with census changes greater than or equal to plus or minus four patients were extracted from the North 6 database. The correlation coefficient between the number of hours the prediction was in error and the net change in census was calculated to be 0.68 with the probability less than .001 that there is no correlation. This clearly suggests that census fluctuation contributes to the prediction errors of the current method. Figure 4.2 plots the paired data. Almost every significant change in census was paired with a large prediction error. 30 · 20 - • · 0 Prediction 10 Errors 0 - • ••• •• · • •• •.... • • • -- · • -10 - • ••• • •• • •• • -20 - •• • • -30 I I -20 -1 0 o 10 20 Net Change in Census Figure 4.2. The correlation between the net change In census and the prediction errors on North 6. 38 However, there were several incidents of large errors that were not related to the net change in census, but it is possible that there were significant turn-overs in the census due to admissions and discharges which were not reflected in the overall net change in census. Thus, it appears that the net change in census is one contributor to errors in prediction; clearly there are additional influencing factors. Generality of the Model If a single model were to serve the units equally well, one would expect similar trends in the relationship between the previous 24 hour acuity and the current acuity. For example, 40% of the previous 24 hour acuity should have a correlation coefficient with the current acuity that is similar on both units. In addition, if the model is adequate, there should be a consistent positive linear relationship between the past and present acuity. The previous 24 hour acuity was treated as an independent variable and tested in a linear regression model for predictive value. For West 5 and North 6, respectively, the correlation coefficients on the original data set were 0.06 and 0.23 with tail probabilities of 0.171152 and 0.001551. In order to visualize the relationship, the predicted acuity of the original data set was plotted against the actual acuity. See Figures 4.3 and 4.4. The North 6 data appears more linear than the West 5 data. This is consistent with the correlation coefficients: although the correlation coefficient on North 6 still only explains less than 25% of the variation. 39 110 100 Evening Acuities 90 • • •• •• • • , • ••' . , • •• •• 80 • 70 • • 60 70 80 90 100 11 40% of the Previous 24 Hour Acuity on West 5 Figure 4.3. The relationship between the previous 24 hour acuities and the evening shift acuities on West 5. 70 •• • 60 • • Evening • • , • • • • Acuities 50 • • •• ••• • • 40 • •• • • • • 30 • • • • • • 20 • • 10 20 30 40 50 60 40% of the Previous 24 Hour Acuity on North 6 Figure 4.4 The relationship between the previous 24 hour acuities and the evening shift acuities on North 6. 70 40 Obviously the model has a higher predictive value for North 6 than for West 5. However, the fluctuations in the census and the workload per nurse that the prediction errors create, support the nurses' claims on North 6 that the model is not good enough to meet their needs. CHAPTER 5 DEVELOPMENT AND EVALUATION OF THE LINEAR REGRESSION MODELS The second question this study addressed was: could a linear model be developed that would improve the acuity predictions? This chapter gives the results of the methods used to formulate the linear models on both units and compares those models to the current model. The Development of the Linear Models Individual variables were correlated with the evening shift acuity to determine the probability that their variation with the afternoon acuity score was random. The data set utilized in the West 5 regression was from February 22, 1986 until March 21, 1986, twenty-eight data items, and for North 6 the data set was from February 12, 1986 until March 1, 1986, thirty-eight data items. Both sets of data utilized in the formulations of the linear regression equations are referred to as the original data sets. The single variable regression r2 values and tail-probabilities for West 5 and North 6 are listed in Tables 5.1 and 5.2, respectively. The variables are ordered according to ascending tail-probability values. Those variables beneath the indicated line were not Table 5.1 Correlation Coefficients of the Independent Variables Regressed on the W es t 5 Evening Shift Acuity Score Variable Tail-Probability IV's 0.390 0.0003 Day acuity 0.276 0.0030 Net change in census 0.270 0.0044 Telemetry 0.200 0.0123 Oxygen 0.203 0.0127 Teaching 0.160 0.0300 Vital signs 0.126 0.0550 Linen changes 0.111 0.0720 Day shift census 0.110 0.0750 Drains 0.107 0.0783* Discharges 0.080 0.1300 Admi ts 0.025 0.4040 Dressings 0.024 0.4080 Medications 0.010 0.5860 *Variables with higher tail probabilities were not considered for further linear regression analysis. 42 Table 5.2 Correlation Coefficients of the Independent Variables Individually Regressed on the North 6 EVening Shift Acuity Score Variable Tail-Probability Day shift acuity 0.549 0.000000 Medications 0.425 0.000006 Cens us 0.386 0.000021 IV's 0.301 0.000262 Vital signs 0.277 0.000505 Oxygen 0.200 0.003858 Net change in census 0.122 0.027165 Teaching 0.087 0.086739 Surgery 0.067 0.106940* Drains 0.065 0.111636 Linen changes 0.028 0.304377 *Variables with higher tail probabilities were not considered for further linear regression analysis. 43 included in further regression models. The list of variables illustrates some differences and similarities between the two 44 nursing units. Census was a good predictor on North 6. This was consistent with the earlier analysis of census fluctuations and prediction errors. Also the number of medications given was a good predictor on North 6 but not on West 5. Because of the extent of potential combinations of variables that could be fed into the linear regression model, correlation between variables was calculated to narrow the focus. Tables 5.3 and 5.4 are the correlation matrices of the variables that were considered for further regressions on West 5 and North 6, respectively. The highly correlated variables, were not usually tested in combination. When two variables were highly correlated with each other, the one that had the higher correlation to the dependent variable was selected for consideration. It can be seen that the net change in census had the lowest overall correlation to all the other variables for both units. After numerous trials, a single linear model was selected for each unit on the following basis: the model having the highest r2 value while the tail probability of each variable remained less than 0.05. The optimal linear regression model found for West 5 was Evening Acuity = -12.95 ,+ (0.66 x IV's) + (1.83 x day shift census) + (2.21 x net change in census) and for North 6 was Evening Acuity = 2.18 + (1.08 x day shift census) + (1.67 x net change in census) + (0.39 x day shift acuity). Table 5.3 The Correlation Matrix of the West 5 Variables Day Day Net Acuity Census Change Teach IV's Tele Linen 02 Vitals Drains ----------------------------------------------------------------------------------------------------------- Day Acuity 1.00 0.45 0.18 0.36 0.48 0.25 0.71 0.29 0.39 0.47 Census 0.45 1.00 -0.37 -0.13 0.16 0.22 0.39 0.04 0.53 0.16 Net Change 0.18 -0.37 1.00 -0.25 0.32 0.04 0.24 0.35 -0.03 0.21 Teach -0.36 -0.13 -0.24 1.00 -0.55 -0.17 -0.20 -0.41 -0.33 -0.42 IV's 0.48 0.16 0.32 0.55 1.00 0.46 0.21 0.31 0.48 0.49 Telemetry 0.25 0.22 0.04 -0.17 0.46 1.00 0.21 0.20 0.07 -0.11 Linen 0.71 0.39 0.24 0.24 -0.20 0.21 1.00 0.31 0.29 0.05 02 0.29 0.04 0.35 -0.41 0.31 0.20 0.31 1.00 -0.00 0.00 Vitals 0.39 0.53 -0.03 -0.32 0.48 0.07 0.28 -0.00 1.00 0.38 Drains 0.25 0.16 0.21 -0.42 0.49 -0.11 0.05 0.00 0.38 1.00 ------------------------------------------------------------------------------ .p,. l.Il Table 5.4 The Correlation Matrix of the North 6 Variables ---------------------------------------------------------------------------- Day Day Net Acuity Meds Census IV's Vitals ~ Change Teach Surgery - Day Acuity 1.00 0.73 0.80 0.61 0.80 0.42 0.00 0.59 0.47 Meds 0.73 1.00 0.58 0.59 0.62 0.63 0.02 0.28 0.38 Day Census 0.80 0.58 1.00 -0.74 0.63 0.22 -0.29 0.63 -0.19 IV's 0.61 0.59 0~74 1.00 -0.45 0.34 -0.12 0.39 -0.04 Vitals 0.80 0.62 0.63 0.45 1.00 0.31 0.08 0.32 0.82 ~ 0.42 0.63 0.22 0.34 0.31 1.00 0.13 -0.13 0.29 Net Change 0.00 0.02 -0.29 -0.12 0.08 0.13 1.00 -0.30 0.23 Teaching 0.59 0.28 0.63 0.38 0.32 -0.13 -0.30 1.00 -0.03 Surgery 0.47 0.38 0.19 0.04 0.82 0.29 0.23 -0.03 1.00 .f::.. 0\ 47 Tables 5.5 and 5.6 give the analysis of the variance for these two equations and Figures 5.1 and 5.2 show the plots of the prediction errors on the original data set for West 5 and North 6 respectively. The standard deviations of the prediction errors for West 5 and North 6 were 4.5 and 6.2 hours respectively. These are considerably less than the current model on the same pieces of data. They were 9.7 and 10.4 hours of nursing care respectively. However, this is not a fair comparison because the original data set was used to formulate the linear models. In other words, the linear models were tailored by this particular set of data and will naturally have a minimal error rate. Therefore, both models were tested on a separate set of data referred to as the test set. Testinfi of the Linear Models In order to test the predictive value of each linear regression equation, a prediction was done on a separate set of data, which was not used in formulating the equation. The West 5 linear regression model was initially tested on data obtained In January. It was noted that all the predictions were very low. Upon reconsideration, it became clear that the number of IVs extracted from the patients' files was unreasonably low. This phenomenon was probably caused by some technical difficulties that occurred in January concerning the storage and retrieval of acuity data. Table 5.5 Analysis of the Linear Regression Variance on the West 5 Original Data Set Variable Constant /IV's //Census //Net Change *r = 0.71 Coefficient -12.95 0.66 1.83 2.21 St.error 18.09 0.25 0.42 0.46 Table 5.6 t-Value -0.7 2.7 4.3 4.8 Analysis of the Linear Regression Variance on the North 6 Original Data Set Variable Constant //Census //Net Change / Day Acuity *r = 0.75 Coefficient 2.18 1.08 1.67 0.39 St. error 4.77 0.35 0.32 0.18 t-Value 0.5 3.1 5.2 2.2 48 Tail-Prob. 0.480998 0.013803 0.000250 0.000007 Tail-Prob. 0.649743 0.004363 0.000011 0.034622 Prediction Errors 20 10 o -10 - • . - 60 49 • ..•. ... • •• • •.• a . •..• • • • I I I I 70 80 90 100 1 ; 0 Predicted Acuities on West 5 Figure 5.1. The prediction errors of the linear regression model on the West 5 original data set. 20 · • 10 -· • •• . •• ••• • ••• •. . • • • • · • • • • • •• •• •• • Prediction Errors 0 -10 - • • · • • -20 I I I I . 20 30 40 50 60 70 Predicted Acuities on North 6 Figure 5.2. The prediction errors of the linear regression model on the North 6 original data set. 50 It was decided to select a later time period to retrieve the test set data. A week in May was chosen because of the easy accessibility of the current patient data file. Figures 5.3 and 5.4 show the prediction errors of the May test set for West 5 and the April test set for North 6, respectively. The mean prediction error on the West 5 test set was 0.3 hours and the standard deviation was 8.8 hours. The mean prediction error for the North 6 test set was 1.17 hours and the standard deviation was 7.6 hours. The durability of the derived equations was tested by combining the original data sets with the test sets and again regressing them against the dependent variables. The data for West 5 was increased to thirty-eight pieces of data and for North 6 it was increased to fifty-nine pieces of data. The prediction model for West 5 was altered considerably, causing significant changes in the predicted acuities. The original equation was altered to: Predicted Acuity = 21.8 + (0.36 x IV's) + (1.26 x census) + (2.22 x net change in census) by the addition of the test set. The r2 value dropped from 0.71 to 0.52. The following example illustrates the alterations caused by the added data set. If the number of IV s were forty, the original equation would contribute 14.4 hours and the new equation would call for 30.4 hours of nursing care; a difference of 16 hours. If the census were forty-six, the original equation would contribute 78.2 Prediction Errors 20 10 o -10 -20 · - · · - · 70 51 • • • • ... • • • • I I I 80 90 100 110 Evening Shift Acuities on West 5 Figure 5.3. The prediction errors of the linear regression model on the West 5 test set data. Prediction Errors 20 10 o -10 -20 - . - . 20 • • • • •• • • • • -. • ..... ~ ~ • •• • •• • I I I . I 30 40 50 60 70 Evening Shift Acuities on North 6 Figure 5.4. The prediction errors of the linear regression model on the North 6 test set data. 52 hours and the new equation would call for 57.9 hours; a difference of 20.3 hours. These changes in the equations indicated that the regression equation for West 5 is very data sensitive. However, there was little change in the equation for North 6 when the test set data were added to the original data. The original equation was altered to: Predicted Acuity = 4.9 + (0.98 x census) + (1.56 x net change in census) + (0.38 x day shift acuity) The r2 value only changed slightly from 0.75 to 0.74. In this instance, if the census were forty-six, the original equation would call for 49.6 hours of nursing care, and the new equation would contribute 45.3 hours; a difference of only 4.3 hours. This durability of the North 6 model reinforces the stronger linearity in the relationship of the independent variables and the evening acuities on North 6. Comparison of the Linear and Current Models A comparison between the linear regression model and the current model was made on the test set data. The prediction errors of the linear models were plotted with those of the current model In order to visualize the two methods together. Figures 5.5 and 5.6 show the plots for West 5 and North 6, respectively. Lines were drawn to compare the incidence of prediction errors that fall outside the plus or minus eight hour criterion. 20 · 10 - Prediction · Errors 0 -10 - · -20 70 • • 0 • 0 0 • 00 ' . ... ".. • 0 0 . 0 I . I . I . 80 90 100 110 Evening Shift Acuities on West 5 • W5 Linear o W5 Current Figure 5.5. The prediction errors of the linear model and the current model on the West 5 test set data. 20 10 Prediction Errors 0 -10 -20 . - - 20 .. 0 • • • 00 0 0 .0 .0 • •• • .... . ... ~o .... 0 • • 0 •o . .... •• • 0 0 0 0 0 0 I . I . I I 30 40 50 60 70 Evening Shift Acuities on North 6 • N6 Linear o N6 Current Figure 5.6. The prediction errors of the linear model and the current model on the North 6 test set data. 53 54 Although the sample SIze was quite small, particularly on West 5, the analysis of the current model on the test set data was consistent with the analysis of the current model using the original data set (see Table 4.1). The means and standard deviations were very similar; both samples illustrate the tendency toward over­prediction on the West 5 evening shift. The linear models had smaller standard deviations on both units, but the minimum and maximum values showed the wide range of prediction errors in both models. The current model's minimum prediction error of -18.7 on West 5 compared to the maximum of 8.6, again illustrated the tendency to overpredict on this shift. Statistical tests were applied to compare the prediction errors of the current and linear models. The paired t-test gave the following results: West 5 North 6 t value 1.89 0.78 Significant at the 5% level >1.85 >1.72 These results would indicate that there is a significant difference between the two models for West 5 but not for North 6. However, the paired t-test assumes an underlying normal distribution of the prediction errors. Since this cannot be fully assumed and the sample sizes were small, the nonparametric Wilcoxon signed rank test was applied to the West 5 data. 55 The sum of the negative ranks and the positive ranks were 9 and 36, respectively. Significance at the 5% level requires values outside <5,40>. The signed rank test would not allow the null hypothesis of no difference between the prediction errors to be rejected. The percentage of times when the first criterion of plus or minus 8 hours could not be satisfied decreases from 52% uSIng the current model to 33% using the linear model on North 6. There IS no change on West 5. CHAPTER 6 DISCUSSION AND CONCLUSION This research project was proposed in order to develop a staffing model that could be easily automated on the computer system and that would be an improvement over the current model. Since the eventual goal at LDS Hospital is to have ongoing access to the patient's current acuity data through computer charting, it was hoped that the acuity data would contain valuable information for predicting staffing needs. Although the results of this study did not produce the sought after model, the inadequacies of the current model were demonstrated. In addition, particular characteristics were identified as being essential to a reliable model. Interpretation of Results Inadequacy of the Current Model Because North 6 was the unit which strongly complained that the current staffing model was inadequate, it is ironic that, on paper, the current model works better for North 6 than for West 5. The relationship between the past 24 hour acuities and the current acuities was somewhat linear on North 6 with a correlation coefficient of 0.23. The mean prediction error on each shift was close to zero on night shift and day shift, and slightly overpredicted 57 on evening shift. The average workload distribution was consistent with the current model. In contrast, West 5 was generally under­predicted on the day shifts and overpredicted on the evening shifts. The correlation coefficient between the previous 24 hour acuity and the current acuity was only .06 and the average distribution of the 24 hour workload was not the same as the model assumes. The mean prediction errors were all larger than North 6's and 56% of the time on the evening shift, they were outside of the eight hour criterion. The application of the second criterion explains the discrepancy between the current model's actual inadequacy and the perceptions of that inadequacy on the two units. The impact per nurse on North 6 is extremely high during those shifts that vary from the averages used in the model. It is a small unit with fewer nurses scheduled per shift and with a large fluctuation in census, workload distribution, and acuity totals. West 5, on the other hand, is a very large unit. It simply does not feel the same impact from the prediction errors because the larger staff can more easily absorb the excess hours of work. Flagle observed that, "the statistical law of large numbers tells us that for any specified probability of available service the necessary reserve capacity is proportional to the square root of the capacity required for average demand." 1 The number of staff needed to maintain the average workload of a unit will increase with the size of the unit. However, the proportion of the square root to the total number will decrease. Therefore, if the unit doesn't have that reserve, the burden per nurse will be less on the larger units. 58 Accuracy of the Acuity Scores The analysis of the claimed hours of nursing care versus the actual hours of nursing care available suggests that the acuity tool has some degree of validity. The mean number of nurses claimed was expected to be close to the number available if the times assigned to the nursing procedures were reasonably accurate. But if these times were over- or underestimated, a consistent pattern of differences between actual and claimed numbers of nurses should be evident. Three of the ten comparisons had an average of the claimed nurses being close to or equal to one full nurse more than the available nurses. The remaining seven claimed less than half a nurse in excess of the number of nurses available. The normal variation in the amount of time it takes for a nurse to perform the required procedures, and the acceptable variation in the assumption of 7.6 hours per nurse are more than enough to account for the small variation in the means. The fact that none of the mean values are less than the actual number of nurses available suggests one of two possibilities: (1) the standard times are higher than the actual time it takes to complete the task, or (2) the nurses generally work more than 7.6 hours in a shift. Interestingly, when the units had fewer staff, the acuities claimed that each nurse did more than 7.6 hours of work but when the units had more staff, there was no significant difference between that claimed and that available. This suggests that the nurses actually do work more than the 7.6 hours when they have fewer staff or that they do the procedures faster than the standard 59 times In order to get their work done. If the standard times were excessive, there should be a consistent pattern of overclaiming the amount of work done. However, in order to verify the cause of this discrepancy, the time and motion study would have to be repeated. The relatively low standard deviations between the claimed and the available numbers of nurses suggests that the tool has some degree of reliability. A wide standard deviation would indicate that the number of nurses on the shift could not be consistently related to the hours of nursing care claimed by the acuity totals. West 5 did have larger standard deviations than North 6, but the highest was only 1.2 nurses. This was not unexpected, because when a unit with more staff has the same degree of error per nurse as a unit with fewer staff, it will add up to a larger error. For example, suppose every nurse on both units works or claims to work an additional 40 minutes, or 0.6 hours. If the unit has 5 nurses, this will total 3.3 hours or 43% of a nurse; but a unit with ten nurses will have an error of 7.2 hours or a full nurse. Therefore, it is expected that West 5 would have wider standard deviations because it is a larger unit and typically staffed with nine or more nurses. The Importance of Census Fluctuation The results of this study contradict the contention that the census and the change in census are not good predictors of nursing workload. 2 ,3 The net change in census between day shift and evening shift, which is not considered in the current model, had a correlation coefficient of 0.68 with the prediction errors on North 6; both census and the net change in census had high correlation 60 coefficients when regressed independently against the evening acuities; both census and net change in census ended up being in the optimal linear regression models found; and on both units, net change in census was the variable most independent relative to all the others studied. It therefore seems that the wide ranges found in the percentage of the workload distribution that each of the three shifts experienced are very likely related to the changes in census. Flagle had pointed out that, "although mere presence as a patient is not a major determinant of workload, it does require a set of activities, a flow of visitors and inquiries, the maintenance of records, communications, and preservation of the nurse-patient relationship. "4 This observation points out that four hours of nursing care for one patient is probably not the same as four hours of nursing care divided between two patients. There is a ... disruption' or ... busy' factor that is not accounted for in the models that assume nursing tasks are performed In a sequential order. The introductory chapter describes the method of staffing previously used in the LDS Hospital that was based on census and census alone. The problem with this method of staffing may not have been with the use of the census, but rather with the lack of a means to update the workload per patient (which has probably increased since the mid-seventies). What the Assessment Tool Revealed It was unfortunate that the results of the nursing assessment tool could not be used directly in the study because the response rate was simply too low, 23% for West 5 and 43% for North 6. However, the evaluations that were filled out proved to be very important In the analysis of the current model. 61 It was noted that the assessment of inadequate and adequate staffing were not at all consistent with the prediction err.ors. For example, the prediction would be fifteen hours short of the actual acuity and the nurse in charge would claim that the unit was adequately staffed. In other examples, the nurses would claim they were significantly short-staffed and the prediction would be ten or more hours beyond the actual acuity. Further investigation revealed that the number of nurses the prediction model would allow was not the same as the number the unit actually had. In other words, the hospital was not fully relying on the model. Some occurrences of understaffing may have been due to lack of available staff and some may have been due to assessments on the part of the head nurses that the tool's prediction was excessive. On the occasions when the number of staff exceeded that which was allowed by the prediction, it is obvious that adjustments were made because of the discrepancies between the actual needs and the model's assessment <of the needs. In addition, this method, like the current model, made no allowances for the professional mix or the level of experience with regard to the personnel staffing the unit. Frequently the cause of inadequate staffing cited on the staffing assessment tool was an incorrect mix of personnel. The flat rate of 7.6 hours per nurse makes no adjustment for nurses who are inexperienced, or who are unfamiliar with the particular unit they have been sent to, or for nurses who are LPNs and need to be covered by an RN. 62 The Usage of Acuity Data in a Linear Model Another finding was that the individual acuity items could not, account for enough of the variability in the acuity scores, and were themselves so highly correlated, that it is probably inefficient to include them in a staffing model. They would prove valuable as a means of classifying the patients if a classification model were selected. Li mi tations There are several limitations common to all staffing studies. The major factor is the subjectivity of determining the specific criterion for classifying a unit as adequately staffed. In addition, the percentage of times that a tool should be expected to meet a set criterion in order to be judged a useful tool, is also subjective. The necessary assumption that the care provided is the care needed underlying all the work measurement tools and the inability to objectively measure quality of care are major criticisms of staffing studies in general. A work measurement tool generally consists of a list of average times for various procedures. These are usually developed through time and motion studies or estimations of times made by professionals, or a combination of both. The tools usually don't account for the professional level of the person doing the task, the differences in the patients who are the recipients of the tasks, or the overlap between the tasks. Testing for validity is limited to 63 repeating the method of development, e.g., the time and motion studies, and testing for reliability depends upon obtaining similar results when the tool is filled out by different individuals. If the tool focuses on the characteristics of the patient or the care being planned for the patient, then a comparison can be made among the nurses filling out the tool. However, a tool like the LDS Hospital acuity tool that lists exactly what has been done for the patient can only be filled out by the one person who took care of that patient. There is no method to compare responses. There were other limitations particular to this study. When the units were understaffed, the total acuity score was probably less than it would have been with adequate staffing. If the actual acuity score was inaccurate, then the measurement of the prediction error was also inaccurate. The nurses did not consistently fill out the staffing evaluations. If the majority of shifts had been evaluated, then those shifts assessed as inadequate could have been removed from the database. An assumption could have been made that any shift evaluated as adequately or more than adequately staffed had a reasonably good acuity total. Unfortunately, with a response rate of 23% on West 5 and only 45% of those responses indicating that the shifts were adequately staffed, there would not have been a sufficient database remaining for evaluation. In addition, inadequate staffing in the shifts of the previous 24 hours may also affect the predictions of the current model. The formulation of the linear models was subject to the errors In the total acuity scores. If part of the variation in the evening 64 shift acuity total was due to error rather than natural variation, the equations would. not entirely reflect the true relationship between the independent variables and the acuity scores. Conclusion The results of this study found the current model for predicting acuities to be inadequate. A model that is based on averages is inappropriate for a system that has highly variable parameters and where every deviation is costly. The ranges In the workload distributions and the correlation between census fluctuations and prediction errors give weight to the argument that the model is not flexible enough to serve the units well. Even though averages are very consistent, the fact of high standard deviations in the prediction errors leaves too many shifts with extreme staffing inadequacies. The discrepancies found between the number of nurses the model predicted and the number actually on the unit support this conclusion. The idea of finding a simple linear model based on easily accessible acuity items is very appealing. The staffing needs could be automated without the need to classify each individual patient. Certain acuity items do have a high correlation to the workload, which is not at all surprising. Unfortunately, they lack the necessary independence (see Chapter 1) and they are not able to explain enough of the variation within the system. Obviously other factors, many of them not measurable, are related to the variability. Wolfe and Young pointed out that a hospital is at the "mercy" of a certain number of chance occurrences inherent to the nature of the 65 environment.5 The nursIng assessments identified some of those random events which transform an adequately staffed shift into one that is inadequately staffed, such as the influx of patients from other units, emergency admissions, and the unexpected occurrences inherent in the hospital environment. The linear models did decrease the prediction errors enough to suggest that fluctuations in census should be considered in a model; and that, if a single model is to be used for the entire hospital, the model must include more parameters than the total acuity score. Recommendations The results of this project led to the following recommend­ation. An additional study needs to be done which includes the following features: (1) establishing a flexible criterion based on the individual unit's capacity to absorb a prediction error; (2) testing on pilot units which are more than adequately staffed for the duration of the study; (3) testing of several different models, such as the current model, a classification model, and a model based on average workload per census; (4) modifying whatever model is chosen to account for the varying levels of educational preparation and experience of the nurses staffing the units. The committment on the part of all the participants to carry out a study for a set length of time and to adequately staff the units being studied is essential. No model should be used on the units while being tested. The staffing needs should be determined by the consensus of several experienced nurses who are not actually working on the units and who have been assigned the task for the duration of the study. _ In addition, the professional level, experience, and area of expertise of the nurses staffing the unit should be integrated into the study. 66 The criterion established needs to reflect the cost of the prediction errors. For example, particular units or shifts using a small number of staff should have stricter criteria because they are less able to absorb the additional hours when short staffed. In addition, a low standard deviation in the prediction errors should be included as a criterion. The purpose of a staffing model is not to predict averages. It must handle the variability that occurs on a shift to shift basis. Notes 1 Charles D. Flagle, Factors Related to Nurse Staffinfi and Problems of Their Measurement, U.S., DHEW Publication, no.(NIH) 73-434. (Report of the May 1972 Conference: Research on Nurse Staffing in Hospitals), p. 30. 2Phyllis Giovannetti, "Understanding Patient Classification Systems, Journal of N ursini Administration (February 1979): p. 4. 67 3 Diane Meyer, "Workload Management System Ensures Stable Nurse-Patient Ratio," Hospitals. J.A.H.A. 52, (March 1, 1978): p. 81. 4 Flagle, p. 27. 5Harvey Wolfe, John P. Young, "Staffing the Nursing Unit Part 1. Controlled Variable Staffing," Nursinfi Research 14, no. 3 (Summer 1965): p. 237. APPENDIX A DEFINITIONS OF SELECTED ACUITY ITEMS 69 LDS HOSPITAL ACUITY TASK DESCRIPTIONS X ___ Means to indicated the frequency of performing the task "Constant" time Vital signs X_ Regular IV X_ IV Hyperalimentation Oral feed minimum Oral feed maximum NG tube Chest tube Routine assessment, q .i.d. vital signs, passing trays and basic set-up of food BP, temp, resp. rate other than qid or post-op Frequency to reflect # of bottles hung plus # of IV sites to main­tain' includes tubing changes, site checks, rate checks at least q 2 hr, bottle change, addition of meds to bottle. Preparation, main tenance, observing for side effects, changing tubing daily. Includes lipid administration. Assist patient with hand washing, tray set-up, open packages, position patient, clean off over-bed table. In addition to oral feed min., total feeding of pt, set-up of kangaroo pump & bag, bag change, calculation & mixing feedings, irrigating tube, checking placement at least q shift. Irrigation, reposition, checking position & patency. Milking tube for patency, measurement of drainage, checking for air leak. Major wound drain treatments X Drainage tubes X __ Supplemental 02 Suctioning X __ Routine and PRN meds X __ Hi-risk meds X_ Glucometer X_ U rine/s tool/drainage tests and specimen collection X __ Complete bathllinen change Partial bath/linen change 70 Frequency reflects # of drains. Includes irrigations, packing, & new colostomy. Reflects # of drains, assessment & emptying of drains, e.g.; hemovacs, sumps. Set-up of equipment including; face mask, tent, headbox, nebulizers, check connections and concentration. Self-explanatory. Preparation and administration of all meds excluding high-risk and/or chemotherapy. Preparation, administration, and close monitoring of effects. Includes meds to be given slow IV push, meds with serious adverse reactions. Self -explanatory Tests for sugar, acetone, protein, guiac, etc., and collection of specimens for lab test. Complete bath, linen change, positioning. Set-up & assist, linen change Other linen changes X __ _ Skin & oral care X __ Isolation, wound, enteric Isolation, protective Dressing change, minor Dressiong change, major Admission Transfer Discharge Surger Pre-op Teaching/emotional support, routine' Teaching/emotional support, difficult Confused/combati ve minor Other than with bath, adding hypothermia blanket. Oral care, pericare, care of incontinence, hair care, back rub, nail care, shave. Self-explanatory. Self-explanatory. 71 Closed, minor or non-infected wounds, central &/or TPN lines. Infected/dehissed wounds with 2-3 step cleansing process. Minor burns (45% body surface area) Admitting assessment, obtaining report, preparation of pt. room. Inter-unit or inter-hospital. Preparation, report, transfer. Discharge summary, preparation of pt & belongings. Patien t preparation, charting. Up to 30 minutes of nursers time. Over 30 minutes of nursers time. Needs reinforcement and frequent orienting (up to 30 Confused/combati ve major Telemetry Post-op care, general anesthesia Post-op care, local anesthesia 72 minutes of nurse's time). Difficult to manage pt., seizuring pt. (more than 30 minutes of nurse's time). Watching monitors, running strips, troubleshooting, contacting doctors. v.s., initial assessment, adjusting pt into room following general anesthesia. v.s., initial assessment, adjusting pt in to room following local anesthesia or procedures such as heart caths or angiograms. APPENDIX B EXAMPLE OF A UNIT ACUITY REPORT 03/12/86.14:21 N:'lME PtiTIENT !iCUIrY REPORT - bl-J 02/1218b ., EVE SHIfT . 12 13 1~ . 15 16 74 HRS R~.r" ID HRS mu In H(S Rtf/, ID 1m RNk!D HRS rur", ID . f+~*f*f++f++t~~ff*ff++fffif+tft+f~ff**!tff*f*f+tf~~t+i~f~f~ft++ffIilf!1111111*++f~t .(3836:~m 3.7.18 1.1.18 1.6.21 1.7.22 1.2.15 (38555~1) 2.8.36 2.3.30 1.2 .15 1.8.23 '. 1.8.23 db ' KMEN (WN 3.9 .51 2.7.35 2.9.38 1.9.25 2.8 .36 (3362901) 3.4 .~4 2.0.7.6 1.8.23 1.8.23 1.~ .18 (3868510) 1.2 .1S (3869898) 0.8 .10 (387111'2) 0.711 .09 (2871787) 0.6 .07 (3373J20) 1.8 .273 (3375408) 2.5 .32 (3876551) 1.1 .57 <:~770b..rv 0.8 .10 (3'377891) 2.3 .30 (3878212) 2.2 .28 (3879~~2) 1.6 .21 (SS79723) 1. 9 .2S 1.1 .H 2.1 .27 1.7 .2'2 1.1 • H (::.379962) 2.2 .2S 0.9 .11 (3880192) 1.0 .13 0.8 .10 (3BS0580) 1,8.23 2.6.~ (3i811l.)) 0.3 • to 1.1.11 ..... F'RAHK LESTER 2.7 .35 - 2.8.36 (1701298) 1.3.23 1.0 .13 (4701~8) 0.7.09 1.6 .21 (:;j7796b) t. 7 .22 <~702023) 3.7 .48 (3876620) 1.1 .1B (3878006) 41. 5 • 19 (3381232) 1.8 • Z3 3.5 .% 1.9 .25 1.2 .15 1.4 .18 1.4 .18 1.~ .1£: 1.4 .J8 ~. 9 .64 2.9 .38 (3881778) 1.30 .17 2.0 .26 ('J'i3b6076) 1.it .18 1.7 .22 (38)6f'.69) 3.0.39 1.7.22 (3380051) 2.5 .32 2.4 .31 (38B0671) 2.2t .28 3.2 .42 .. ' It'''A..trqUf: BaLI 2.1 .31 <::281 if'.6) 3.2 • ~2 (3SS1S15) 1.8 .23 2. 6 ~34 1.3 .17 1.3 .17 1.5 .19 ~.4 /57 2.9 .38 1.3 .17 1.0 .13 2.7 .35 1 .. 9 .25 0.9 .11 2.6 .34 1.Oe .13 (3882776) 2.0 .26 , LfilH ~S11 2. 7 e .3S 2.0 .26 0.7 .09 1.7 .22 3.5 .~b 1.4 .18 1.5 .19 1.1 .18 1.3 .17 1.0 ,,13 2.1 .27 . 3.1 .1tO \~/lP~~~) 1. 7t .22 111~III!l!I!IIIIII!III!IfIIIIIIIIIIII§IIIIIIIIII~IIfll11111!llf!IIIII!Jllllllllllf!IIIfI!Xlllllllllllffl SHIFT TOlliLS 45.6 45.7 ~5.9 37.3 28.6 ~~/PT"SHIFT D;W TOf!"d..3 5.87 5.89 5.96 ".31 3.69 1.98 (2::) 1.82 (25) 2.18 (21) 1.96 (19) 1.78 (16) 127.0 122 .. 9 118.6 92.5 76.9 16.34 15.82 15.31 11.89 9.89 HR/Pr"MY (:;UI) ~. 98 4. 78 ~. 93 1.50 4. b3 HVPl"DP.Y (avg) 5 .. 08 (25) 4.91 (25) 4.94 (24) 4.62 (20) ~.OO (l6) MINUTES TO BE F.E?ORTED TO TRlil'l1JiCT1~ FILE (This rr09r~~ doe~ not send trlns~ction dlta.) APPENDIX C THE STAFFING ASSESSMENT TOOL 76 Unit Staffing Evaluation Questionnaire Date (mo/day/yr) _____________ for nights, give date at end of shift) Shift day __ evening __ night __ _ N arne of the two people completing this form: Staff on Shift Giving Patient Care Number RN's LPN's Aides Students Ward Clerk 1. In general, did you feel the staffing for this shift was: _____ nadequate _____ Adeq uate ___ More than adequate (could have used fewer nurses) 2. If inadequate which one/s of the following factors seemed to best describe the situation? Above average number of patients needing extensive nursing care, assistance or surveillance. Not enough staff on duty: (please circle) a)too few scheduled b )scheduled but not on duty c )transferred to another unit Census fluctuations { e.g. above average number of admits this shift or previous shift, transfers or discharges } . Not an optimal mix of personnel skill levels __ Other (please describe) _______________ _ 3. In your judgement, if staffing was inadequate what would have best helped the situation? Part of Shift Full Shift One additional RN Two additional RN's One additional LPN Two additional LPN's A ward clerk Other (please describe) 4. If there were any special circumstances that occurred during your shift, like a patient needing to be specialed or someone getting sick in the middle of a shift, please describe below. 77 5. If you feel that all the patient care reqtiiren1ents could not be met due to a staff shortage, please check the area of care that was less than satisfactory for a majority of the patients. Be sure to consider factors such as safety, accuracy, patient comfort, and agreement with medical and nursing goals. __ Basic Hygiene __ Basic feeding and toileting ___ Mo bili ty __ Meds & IV's __ Vital signs ___ Communication (teaching, orienting, comfort) __ Special procedures (drsgs, irrigations, cath care, etc.) __ Observation of patients __ Implementation of new orders without undue delay 78 APPENDIX D WRITTEN COMMUNICATIONS WITH THE NURSES ON WEST 5 AND NORTH 6 February 10, 1986 Dear North 6 Nurses . I'm certain not all of you were able to attend the recent staff meeting. I am a graduate student in the biophysics department and I'm doing a study on the nursing acuIties. I have asked your floor to participate and you have generously agreed to help. I'm requiring a small amount of paper work. Please read the following and call me if there are any questions. Purpose: to assess whether the floors are being adequately staffed and if the acuities really reflect the patients' needs. Your Input: at the end of a shift, the charge nurse fills out a form telling me if, in her/his judgment, the staffing was adequate. If not, she/he tells me what additions would have been helpful. The forms are placed in an envelop provided in your dressing room. Please Note: If you forget to fill the form out but remember several days later, go ahead and fill it out. They do not have to be on time. If a form is never filled out for a particular shift, I will try to make the assumption that the staffing was adequate. You will notice that there is a place for two nurses to sign. One of those nurses should be the person in charge. It's nice if she has someone to confer with her decision but it's not absolutely essential. Let me know if you have problems, questions, concerns. For a while I will call at the end of each shift to remind the person in charge. Sincerely, Julia Rossi RN 363-XXXX any time ext. 2XXX (between 1pm and 4:30) 80 81 February 4, 1986 WEST 5 NURSES PLEASE READ!!!!!!!!!!!!!!!!!!!!!!!!! Please remember to fill out the staffing evaluation form if you are inadequately staffed. It is not essential that you fill the form out on the day you worked the shift. For example, if you were extremely busy on a particular shift and forgot to fill out the form, do it when you get the chance even if it's several days later. Don't worry if you can't remember exactly how many RN's or LPN's were on the shift. I check your assignment sheets for every shift. Just tell me why you think you were over- or understaffed and what you would have liked added or subtracted. Please understand that when you fail to inform me that a shift was inadequately staffed, I have to presume that the number of nurses you had was correct for that level of acuity. I urge you to participate in this study and to be supportive of nurses assessing the nursing needs of patients. It really is important. Please call me (including night shift) and I will be glad to talk to you about acuities and why we need to do them. Thank you, Julia Rossi, R.N. LI1ERA TORE CI1ED Aydelotte, Myrtle K. Nurse StaffingMethodology: A Review and Critique of Selected Literature. U.S., D.H.E.W. Division of Nursing, no. (NIH) 73-433. Abdellah, Faye G. and Eugene Levine, Better Patient Care Through Nursing Research. New York: The Macmillan Company, 1965. Connor, R.J. "A Hospital Inpatient Classification System. II Ph.D. diss., The Johns Hopkins Univ., 1960). Flagle, Charles D. Factors Related to Nurse Staffing and Problems of Their Measurement. U.S., DREW Publication, no. (NIH) 73 -434 (Report of the May 1972 Conference: Research on Nurse Staffing in Hospitals). Giovannetti, Phyllis. "Understanding Patient Classification Systems." Journal of Nursing Administration (February 1979): 5-9. Griffith, John R. Impact of Administrative and Cost Factors on Nurse Staffing, U.S., DHEW Publication, no.(NIH) 73-434. (Report of the May 1972 Conference: Research on Nurse Staffing in Hospi tals). Grimaldi, Paul L., and Julie A. Micheletti. "RIMs & the Cost of Nursing Care." Nursing Management 13, no.12 (December 1982): 12-22. Hartigan, John A. LIDA. computer sofware. Yale University, Department of Statistics, 1983. Disk. Kiley, Marylou, et al. "Computerized Nursing Information systems (NIS)." Nursing Management 14, (July 1983): 26-29. Meyer, Diane. "Workload Management System Ensures Stable Nurse-Patient Ratio." Hospitals 52, (March 1, 1978): 81-85. Murphy, Laurel N., et al. Methods for Studying Nurse Staffing in a Patient Unit. U.S., DHEW, no. HRA 78-73, May 1978. Pryor, T. Allan, et al. "The HELP System." Journal of Medical Systems 7, no. 2 (1983): p. 83-87. Trofino, J. "A Reality Based System for Pricing Nursing Service." Nursing Management 19, (January 1986): 19-23. Vanderzee, Holly, and George Glusko. "DRGs, Variable Pricing, and Budgeting for Nursing Services." Journal of Nursing Administration. (May 1984): 11-14. Vaughan, Robert G., and Vernon MacLeod. "Nurse Staffing Studies: No Need to Reinvent the Wheel." Journal of Nursing Administration, (March 1980): 9-15. Williams, Margaret A. "Quantification of Direct Nursing Care Activities. n Journal of Nursing Administration 7, (October 1977) 15-18. 83 Wolfe, Harvey, and John P. Young. "Staffing the Nursing Unit Part 1. Controlled Variable Staffing." Nursing Research 14, no.3 (Summer 1965): 236-43.