Myocardial Infarction Mortality in Rural and Urban Hospitals: Rethinking Measures of Quality of Care

We would like to thank Dr.’s Longenecker and Oliver for their comments and interests in our study. We interpret Dr. Longenecker’s comments in support of our work as indicating some degree of face validity to our findings by someone with significant expertise in the practice and training of physicians for rural communities.

Dr. Oliver raises a thoughtful question of whether there could be other confounding measures still unaccounted for in our analysis. If the spatial autocorrelation truly correlates directly with myocardial infarction mortality and not with choice of hospital, it would be a variable that violates an assumption of the instrumental variable technique. Our results in Table 4 and the Overidentifying Restrictions Tests in Table 5 show that even if other unmeasured confounders such as spatial autocorrelation exist, they do not affect the estimations of coefficients. If there were a strong spatial autocorrelation variable (even though it does not directly affect our outcome), we would have a better model if we were able to control for it. However, instrumental variable models adjusting for autocorrelation will have wider confidence intervals for coefficients than unadjusted models; in other words, we would have an even larger p-value indicating no difference in mortality rates. Thus, we believe that we would draw the same conclusion if we adjusted for spatial autocorrelation.

We agree completely with Dr. Oliver's plea to continue to improve risk adjustment models that relate to important outcomes such as mortality.

Competing interests:   None declared

Patients with acute myocardial infarctions treated in rural hospitals have a higher mortality rate than those treated in urban hospitals. As noted by James, Li, and Ward in their report in the current issue of the Annals of Family Medicine,(1) the conventional wisdom is that this disparity is a result of poorer quality of care in the rural hospitals owing to their having lower volumes of patients with AMI than the urban hospitals. Patient volume, in these studies, is used as a proxy measure of institutional experience and skill. James et al convincingly demonstrate that the finding of increased mortality from AMI in rural hospitals in prior studies may result from flawed modeling, rather than representing differences in the quality of care in the two settings.

James and his colleagues hypothesize that previous models of hospital mortality from AMI may have been biased by unmeasured confounders, and they make innovative use of econometric methodology to adjust for these hypothesized confounders. They develop instrumental variables, grouping patients based upon their distance to the nearest urban hospital.

A cardinal assumption made by health services and primary-care researchers who conduct spatial analyses of disease and treatment outcomes is that the location of research subjects is associated with these outcomes, presumably through unmeasured – and even unmeasurable – risk factors that also are tied to that location. Researchers using spatial analytic measures adjust for these unmeasured confounders by including terms in their models that account for spatial autocorrelation.

James et al have carried out an analogous adjustment with the use of instrumental variables based on geographic distance from the nearest urban hospital. Their model performs better statistically than the traditional analytic models that use risk adjustment models based upon a limited number of patient characteristics. However, the model presented by James et al does not include any adjustment for spatial autocorrelation. It is possible that further confounding factors could be adjusted for in such a manner, improving the model’s performance.

Nevertheless, the finding by James et al that mortality from AMI in rural hospitals is not higher than that in urban hospitals after controlling for unmeasured confounding factors underscores the importance of reassessing the measures used to evaluate the quality of care provided by hospitals. Current traditional risk-adjustment methods are not sufficiently controlling for factors affecting mortality risk. Other studies have argued for improved risk-adjustment models.(2-4) James et al provide another innovative approach to accurate assessment of quality of care.

1. James PA, Li P, Ward MM. Myocardial Infarction Mortality in Rural and Urban Hospitals: Rethinking Measures of Quality of Care. Ann Fam Med 2007;5:105-111.

2. Pine M, Jordan HS, Elixhauser A, Fry DE, Hoaglin DC, Jones B, Meimban R, Warner D, Gonzales J. Enhancement of Claims Data to Improve Risk Adjustment of Hospital Mortality. JAMA 2007;297:71-76.

3. Stukenborg GJ, Wagner DP, Harrell FE, Oliver MN, Heim SW, Price AL, Kim C, Wolf A, Connors AF. Present-at-admission Diagnoses Improve Mortality Risk Adjustment Among Acute Myocardial Infarction Patients. J Clin Epidemiol 2007;60:142-154.

4. Stukenborg GJ, Wagner DP, Harrell FE, Oliver MN, Kilbridge KL, Lyman JA, Einbinder J, Connors AF. Present-At-Admission Diagnoses Improve Mortality Risk-Adjustment and Allow More Accurate Assessment of the Relationship Between Volume of Lung Cancer Operations and Mortality Risk. Surgery 2005;138:498-507.

Competing interests:   None declared