Four points warrant further discussion. First, our study has a number of strengths. AHEAD is a nationally representative sample of 7,447 individuals who were 70 years old or older when they completed their baseline interviews in 1993–1994. We linked an extensive array of possible risk factors (both traditional and novel) obtained from the baseline interview data for 5,511 of the AHEAD participants to their Medicare claims for up to 12 years of post-baseline surveillance, and we used propensity score methods to adjust for potential selection bias in the analytic sample. From the claims data we constructed a time-dependent (dynamic) post-baseline "health shock" (recent hospitalization) measure and included it in our multivariable hazards models. Two different approaches for stroke case-identification were considered, one that emphasized sensitivity and one that emphasized specificity, and with both approaches we replicated our analyses after excluding participants who self-reported pre-baseline stroke histories. Among the 5,511 participants, 6.8% and 9.9% suffered a post-baseline stroke (recurrent or first-ever), depending on whether the high specificity or high sensitivity (respectively) case-identification approach was used.
The second point that warrants discussion involves the importance of the dynamic health shock marker , which was measured indirectly using the time-dependent recent hospitalization indicator. This effect was quite large and did not mediate the effects of the baseline stroke risk factors. It captures the transition period when older adults are especially vulnerable to adverse effects associated with both their underlying health shock (i.e., the reasons for their recent hospital admission) and the consequences of their treatments, especially in fragmented health care delivery systems [39–43]. When calibrated at 7 days, which is when the effect size peaked, the health shock measure increased the risk of stroke by about 200–480% (depending on the case-identification approach and restriction to first-ever strokes), and substantially improved model fit. This suggests that post-discharge planning and monitoring for a week or so following hospital discharge for something other than a stroke might be fruitful and might potentially reduce the risk of subsequent stroke during this transition period. It would seem prudent at this point to design and evaluate a pilot, short-term, post-discharge planning and monitoring intervention study, consistent with the recent work and suggestions of Coleman and colleagues [45–47].
Although the introduction of the health shock (i.e., recent hospitalization) measure is a very promising development that underscores the need to shift from static to dynamic risk modeling approaches , further research is needed. That research should explore the health shock measure in order to clarify what the underlying etiologic mechanism(s) might be. Such research should include whether restrictions to surgical vs. medical admissions, shorter vs. longer stays, or other decompositions would identify particular hospitalization subsets that pose the greatest risks for subsequent stroke.
The third discussion point involves the identification of what we did find; that is, the static baseline risk factors and the magnitudes of their risks. Regardless of the case-identification approach used (high sensitivity vs. high specificity) and whether first-ever and recurrent strokes or just first-ever strokes were considered, our risk estimates were remarkably consistent. The greatest risks involved increased age, individuals who were widowed or never married, participants living in multi-story buildings, individuals reporting a baseline history of diabetes, hypertension, or stroke, and participants who reported difficulty picking up a dime, refused to answer the delayed word recall test, or who had poor cognition scores. With two exceptions, what we found is generally consistent with the extant literature [1–7, 10–20].
The two exceptions involve the risks associated with multi-dwelling residence and the protective effect of angina. The significant stroke risk associated with multi-story residential dwellings vs. single-story residential dwellings has not previously been reported in the literature. This 40% increased risk was also remarkably consistent regardless of the case-identification approach and the type of strokes considered. Because the point estimates (AHRs) obtained for the other dwelling unit contrasts approximate unity (vs. AHRs ≥ 1.39 for this specific dwelling unit contrast), these differences are not due to insufficient statistical power. Furthermore, because the same differential dwelling unit contrast pattern was observed among the crude HRs, these results are not due to statistical over-adjustment or to multicollinearity. We believe that the increased risk of multi-story residential dwelling reflects the greater physical, social, and psychological burdens faced by older adults in those settings. Although this interpretation is consistent with the literature on congested living and stress [48–50], replication of these results using other nationally representative samples and similar residential dwelling contrasts will be necessary to move this interpretation beyond post hoc speculation.
The protective effect observed for angina is less straightforward for two reasons. First, the independent effect is only manifest with the high sensitivity definition when both those with first-ever and recurrent strokes are considered. Second, the statistically significant independent effect of angina is protective, while it's statistically significant crude effect placed participants at risk. Based on these facts, we assume that the observed protective effect of angina is a statistical artifact, possibly due to statistical over-adjustment.
Our fourth discussion point involves what we did not find. Specifically, we did not observe elevated risks for men, minorities, geographic region, socioeconomic gradients, or obese participants, despite the fact that these are generally reported in the literature [1–7, 10–20]. Determining why elevated risks for these factors were not observed in the AHEAD is beyond the scope of the present study. Nonetheless, we are convinced that this is not an artifact of statistical over-adjustment because these factors exhibited no crude associations with post-baseline strokes. We also expect that given our sample size and the distributions of these risk factors in the AHEAD, these non-findings due not result from insufficient statistical power. Further research, however, using other nationally representative samples will be necessary to resolve these issues.
Finally, in concluding this article, we note the three major limitations of our study. First, family history, dyslipidemia, and asymptomatic carotid bruit were not available for inclusion in the analysis. Second, although the AHEAD is rich in self-reported data and linked to Medicare claims for more than a decade, detailed clinical histories were not available, restricting our study to an epidemiologic vs. etiologic analysis. Lastly, we relied solely on baseline (i.e., static) risk factors from the AHEAD self-reports, even though several of them (such as ADLs, IADLs, and self-rated health) were repeated at most follow-ups. Inclusion of those repeated self-reports, however, would have created numerous additional complexities involving missing data, selection bias, and correlated error structures.