This was a retrospective case series based on the electronic health records (EHR) of adult patients served by Essentia Health East in Duluth, MN. Eligible patients were aged 40–70, had an Essentia Health East primary care provider and received care within Essentia Health East facilities between July 1, 2004 and September 30, 2009. This study was reviewed and approved by the Essentia Health Institutional Review Board.
Study population
To be included in the study, eligible patients were required to have (1) been initially prescribed ACEI or ARB medications between January 1, 2005 and December 31, 2008, without a documented discontinuation for at least 6 months; (2) a diagnosis of DM, CHF, and/or HTN; (3) documentation of baseline Hgb level (12 months before to 10 days after initiation of ACEI or ARB medication) and Hgb level during the follow up period (3 to 12 months after initiation of medication); and (4) baseline GFR or data needed to compute GFR (12 months before to 30 days after initiation of medication). ACEI use has been found to be associated with a decrease in erythropoietin concentrations after as little as 28 days; [12] in this study follow up Hgb was assessed in the 3 to12 month window following the initiation of therapy. There were 7042 eligible patients who were between 40 and 70 years of age at the time of treatment initiation.
Patients were excluded from the study if their medical record included evidence of (1) underlying conditions associated with anemia (hemoglobinopathy, bleeding disorder, chronic inflammatory disease, or treatment with vitamin B12 or folate) during the time from 6 months before initiation of ACEI or ARB until the end of the follow up period; or (2) other conditions or treatments that might affect Hgb level during the follow up period (blood transfusion, pregnancy, hemodialysis, malignancy requiring chemotherapy or radiotherapy, hospitalization, or treatment with EPO).
Following these exclusions, 5613 patients remained eligible for the study. Of these, 5104 met inclusion criteria (2), above, and 839 met both (2) and (3). The study population consists of the 701 patients who met all four inclusion criteria.
For eligible subjects data was extracted from the EHR: age, sex, class of medication (ACEI or ARB), underlying diagnoses (DM, CHF, and/or HTN), baseline Hgb and GFR, and follow up Hgb. If more than one Hgb report was recorded during the follow up period (3 to 12 months after the initiation of medication), the measure closest to 3 months was used. Chronic kidney disease (CKD) was defined as GFR < 60 mL/min/1.73 m [2].
Data analysis
Preliminary analysis of the ACEI and ARB groups at baseline identified differences between the treatment groups with regard to sex; prevalence of HTN, CHF, and CKD; and baseline Hgb; although not all differences were significant at the 0.05 level. In addition, it was noted that both the rate of prescription of ARB vs. ACEI and the follow up Hgb decreased slightly over the four year period. Accordingly, the fractional number of years from January 1, 2005 until treatment was initiated was included as a variable in the analysis. The investigators recognized that in clinical practice, provider-selected treatment is rarely, if ever, random. The preferential ordering of ACEI or ARB for subpopulations defined by sex and co-morbidities was not due to specific treatment guidelines or standards of clinical practice. Accordingly, the treatment effect of ARB relative to ACEI was considered to be truly confounded by the covariates for which treatment group differences had been identified. It was thus necessary to consider the associations between the baseline covariates and both follow up Hgb and providers’ treatment choices, in order to produce an unbiased estimate of the treatment effect.
We estimated the effect of the two treatments on follow up Hgb with the use of the doubly robust (DR) semiparametric efficient estimator [13, 14]. This method produces an estimate of the causal effect of ARB vs. ACEI on follow up Hgb by simultaneously incorporating (1) the chance that a given subject would receive ACEI or ARB, given their baseline characteristics (the propensity score), and (2) the effects of baseline characteristics (covariates) upon the outcome of interest (follow up Hgb). This approach to data analysis is “doubly robust” in the sense that the model produces an unbiased estimate of the causal effect if either (1) the effect of baseline covariates on the outcome or (2) the propensity score is correct. A complete-case ANCOVA was conducted. All analyses were completed using SAS ver. 9.2 and the SAS macro, %drmacro, described by Funk et al. [15] was used to produce the estimated causal effect and asymptotic confidence interval (CI). The bias-corrected and accelerated (BCa) bootstrap CI [16] of the causal effect was constructed for comparison with the asymptotic CI.