| OCR Text |
Show favor of his/her staying at the hospital until s/he was better. 8. If you knew that the patient's family was anxious to take him/her home and provide outpatient therapy for his/her condition. 9. If you knew that a suitable specialist was available in town for consultation. 10. If you knew that no suitable specialist was available for 300 miles. 11. If you knew that a suitable specialist was available by 2-way video conferencing (telemedicine). Ratings of the influence of non-medical factors were on a 1-5 scale (1=much less likely to admit, 2=less likely to admit, 3=neither more nor less likely to admit, 4=more likely to admit, 5=much more likely to admit). Pro-outpatient factors were defined as those which were likely to keep patients out of the hospital. Pro-inpatient factors were likely to increase propensity to admit. The nature of some factors was not immediately obvious and was determined by factor analysis (see statistical methods). Respondents received the initial survey, a postcard reminder, and a second mailed survey if there was no response. A subgroup agreed to receive the survey an additional time to estimate test-retest reliability. Surveys were sent to120 rural and 80 urban physicians, which was anticipated to result in 60 subjects per group, based on the results of the pilot study. This final sample was anticipated to have 90% power to detect differences of 0.66 standard deviations between urban and rural physicians in reported likelihood to admit. Statistical methods Frequency distributions of responses were reviewed to examine variability or consensus of response. Means and standard deviations of responses were calculated for urban vs. rural physicians. Demographics were compared by chi-squared tests; Fisher's Exact test; and t-tests or their non-parametric analogs if the data appeared inconsistent with the normality assumption. In order to summarize the influence of non-medical factors for ACS admissions as a group, "pro-outpatient" and "pro-inpatient" summary scales were produced by summing physicians' ratings over all diagnostic scenarios and the three non-medical factors of family support, proximity to care, and family preference regarding hospitalization. Factor analysis with varimax rotation demonstrated one pro-outpatient and one pro-inpatient underlying factor, explaining 67% of the variance. This supports the validity of these constructs (Hair et al., 1995). Internal consistency based on Cronbach's alpha was 0.84 for the pro-outpatient summary scale and 0.72 for the pro-inpatient summary scale. Test-retest reliability of each of the pro-outpatient and pro-inpatient summary scales was 0.77. Univariate and multivariate analysis of variance methods, including Hotelling's T-squared, were used to test the following null hypotheses when the normality assumption appeared to be satisfied. Null hypotheses: 1) On average, physicians were neither more nor less likely to admit a patient when non-medical factors came into play vs. when they did not. 2) Physicians were no more nor less likely to admit for one non-medical factor than for its opposite. 3) Average ratings of the same non-medical factor and summary scales did not differ by diagnosis. 4) Average ratings and summary scales did not differ for urban vs. rural/frontier physicians. The Mantel-Haenszel chi-squared test of trend was used to test individual survey items on a 1 -5 scale. The age and gender of non-respondents were compared to those of respondents. In addition, the potential effect of non-response due to disagreement with the study hypothesis was considered by a sensitivity analysis. This analysis examined whether the observed findings would have persisted if there had been systematically biased non-response in P % of the planned sample, for P ranging from 0% to 60%. The effect size and p-value were calculated under the scenario in which 120 rural and 80 urban physicians had returned the survey, with P% responding that non-medical factors made absolutely no difference in their admission decisions, and the rest responding with exactly the same means and standard deviations as observed. Results 66 physicians responded, 3 refused, and 120 did not respond. Six respondents said that they do not admit patients. 23 respondents agreed to retake the survey at a later date to provide an estimate of test-retest reliability, and 20 returned the retest survey. Table 1 compares demographic characteristics of urban and rural/frontier respondents. Respondents were primarily male, reflecting Utah physician demographics, and all were Euro-American. 46% of urban respondents characterized their specialty as family medicine, compared to 83% in rural/frontier Utah (p=0.008). Thirty hospitals out of 40 general, acute care hospitals in Utah were represented, in 21 out of 26 counties with 21 |
