In response: Incident depression increases medical utilization in Medicaid patients with hypertension

Response to: Breunig I. Incident depression increases medical utilization in Medicaid patients with hypertension. Expert Rev Cardiovasc Ther 2015;13(1):17-18

We thank JM Jerrell and A Tripathi for their insightful comments and the opportunity to clarify some points from our work [1]. The research design was a pre-post comparison with control group or ‘difference-in-difference’ approach, applied to medical encounters and prescription drug event files for Maryland State Medicaid beneficiaries between 2005 and 2010. We imposed a 2-year period of uninterrupted Medicaid coverage for a complete picture of medical encounters and prescription drug use for each patient. Whether due to ‘churning’ – the involuntary movement of consumers from one health plan or system of coverage to another – or otherwise, we found that imposing a longer uninterrupted period of complete observation would have limited the number of patients observed with newly diagnosed depression and meeting the inclusion criteria, thereby limiting our ability to derive meaningful conclusions and inform the discussion on the relationship between depression and cardiovascular disease (CVD) (in the context of informing delivery of care). We wished to avoid wasting such an opportunity since our study is one of the few to rigorously examine the burden of onset of depression in patients with hypertension, an important risk factor for the more serious manifestations of CVD.

Our study examined the association between onset of depression and frequency of encounters in multiple settings (i.e., inpatient, hospital outpatient, emergency room, physician and mental care) rather than overall encounters only. While one alternative to the pre-post design that we employed would have been an interrupted time series design, using no more than 24 monthly periods to calculate the difference scores, we found that the frequency of inpatient, hospital outpatient and emergency room encounters among our depression and matched-control cohorts in any single month would have been too small to derive meaningful conclusions. This is a common problem that dictates using fewer and longer intervals. Thus, we imposed a 12-month pre-post study design. Our results inform future studies that aim to look at longer term outcomes.

Year 1 utilization was indeed controlled for in the regression analysis. Our difference-in-difference approach was consistent with the second of the two ‘preferred methods’ discussed. We used the difference between year 1 and year 2 utilization as the dependent variable. We did not include baseline (year 1) utilization on the right hand side of the equations, nor did we include other baseline patient risk factors on the right hand side. Instead, we applied a ‘first-difference’ transformation of all data in order to ‘eliminate systematic bias, regression to the mean and reduce error variance.’ First-differencing eliminates the effect of patient risk factors that remained constant across the two time periods. Constant patient factors include observable ones (e.g., demographics or baseline utilization) as well as risk factors that are unidentifiable using Medicaid encounters data (e.g., clinical history). It uses the change in observable risk factors that do change across time as covariates (e.g., hypertension treatment or comorbidity status). Each of the change outcomes happened to be normally distributed. As a result, we found that the ordinary least squares regression estimator provided the best model fit, as opposed to the other distributions one might consider under a general linear model framework.

We relied upon previous literature on cardiovascular and mental health to identify the baseline characteristics (measurable with Medicaid data) used to propensity-score match the hypertensive–depression cohort to hypertensive control patients. These are listed in a table of descriptive statistics included in the paper. As described in the article, standard diagnostic steps for propensity-score matching were employed. One limitation of our overall propensity-score matching and first-differencing approach was that it may not be tractable for clinicians to compare the case-mix of the research cohort employed to their own caseloads, beyond matching the overall case-mix outlined in the description of the treatment and matched cohorts.

Our results contribute to the understanding of the added burden of depression on hypertensive cardiovascular disease. We hope that more researchers will rigorously examine the burden of depression in patients with hypertension, perhaps with clinical and administrative data collected intentionally, systematically and over a long period of time. Only through such effort will it be possible to fully understand how depression might be a significant barrier to managing hypertension, other CVD and their manifestations. As acknowledged in the article, the link between depression and hypertension severity may be epidemiological and not necessarily causal. We believe that our results will help elucidate the relationship between these two burdensome diseases and contribute to future research.

Financial & competing interests disclosure

The author has no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.