OCR Text |
Show variation measurable for non-medical factors in the CHF and pneumonia scenarios. Pro-outpatient non-medical factors of family support and family preference not to hospitalize influenced both urban and rural physicians differently for different diagnoses. Pro-inpatient non-medical factors of isolation from family, distance from care and family preference to hospitalize had similar influences for each of the three scenarios, and greater overall influence in rural areas. This leads to the speculation that non-medical issues of access, which are characteristic of practice in non-urban areas, may be detectable in analyses of variation in ACSC hospitalization rates. Whether these issues can be handled most cost-effectively by hospitalization or other solutions may vary. Daily or twice-daily home health visits may be effective in urban areas, but costs of travel may limit feasibility in rural/frontier areas. Remote video observation may be feasible for some conditions but would incur equipment and infrastructure costs. Admission to the equivalent of a skilled nursing facility or 24-hour observational unit may be another approach to consider for equivocal cases of ACSCs, prior to formal hospital admission. Bindman et al. (1995) previously developed surveys of physician practice styles and patient attitudes in order to model hospitalizations for ACSCs in California. Komaromy et al. (1996) found significantly more self-reported practice variation of physicians between zip codes than within zip codes in urban California. Physician practice style was correlated with local hospitalization rates for ACSCs, but the correlation was nonsignificant when socioeconomic status of the local area was taken into account. Komaromy et al. (1996) suggested that, to the degree to which the pro-inpatient factors of homelessness, substance abuse, or other non-medical factors are associated with low income areas in urban California, physicians may have been responding by hospitalizing vulnerable patients. The present study identified pro-inpatient factors relevant to the rural/frontier Rocky Mountain west, including distance from care, inadequate family support at home, and family wish to hospitalize. Whether these are correlated with local ACSC hospitalization rates would be a subject of further research. This study shows no evidence of important differences between urban and rural/frontier physicians' admission styles when pro-outpatient non-medical factors come into play. It can not rule them out, because the study was limited in its statistical power due to a 30% response rate. In the absence of non-response bias, there was 90% power to detect an effect size (Cohen, 1988) of 0.95 standard deviations. The difference between urban and rural/frontier physicians' admissions styles in the presence of pro-outpatient non-medical factors constituted an effect size of -0.26 standard deviations, which did not achieve statistical significance. This effect was less than half the magnitude of the observed difference in the presence of pro-inpatient non-medical factors. Thus, if there really is such a difference in the population, this study offers no evidence that it is large. Another potential effect of a low response rate is statistical bias, i.e., if nonrespondents' propensities to admit differed in some systematic way from that of respondents' propensities to admit. Sobal et al. (1990) and Parsons et al. (1994) have discussed low physician response rates. They argue that many physicians fail to respond because they are simply too busy, regardless of the survey topic. Respondents in the present study's pilot phase reported that in many such cases, receptionists have instructions to discard such mailings. Sobal et. al.(1990) argue that when physicians provide "role-specific answers representative of others in that role", these characteristically low physician response rates are unlikely to bias the results. Some physicians are over-surveyed. One non-responder to the present study reported that he had received six other surveys in the past month, ours being the seventh. But if the physician looked at the survey at all, it would be subject to content-related biases. For example, the physician who felt that his or her decision to hospitalize would not have been changed by any non-medical factors might choose not to return the survey, rather than repetitively mark 3's on every item. If all nonresponding subjects believed that non-medical factors would not alter their admissions decisions in these scenarios at all, a sensitivity analysis indicated that the observed response to negative non-medical factors would be diluted to the point of statistical non-significance. However, if up to 30% of the planned sample failed to respond for this reason, the finding of rural-urban differences in response to negative non-medical factors would still have been statistically significant. Nonresponders to this study were 95% male, and Euro-American, as were responders. Nonresponders were of average age 47 as compared to 45 among responders (p=0.17 by 2-sample t-test). Thus, although non-respondent 23 |