Comparative effectiveness research using claims data: meticulous methods don’t solve old problems

This commentary refers to ‘Sodium-glucose cotransporter 2 inhibitors vs. sitagliptin in heart failure and type 2 diabetes: an observational cohort study’, by E.L. Fu et al., https://doi.org/10.1093/eurheartj/ehad273.

The paper by Fu et al.¹ assessed the effectiveness of sodium–glucose cotransporter 2 inhibitors (SGLT2i) compared to sitagliptin in Medicare beneficiaries who have type 2 diabetes mellitus and heart failure (HF). The initiation of SGLT2i compared to sitagliptin was associated with a lower all-cause mortality [adjusted HR 0.70, 95% CI 0.63–0.78], hospitalizations for HF [adjusted HR 0.64, 95% CI 0.58–0.70], and urgent visits requiring intravenous diuretics [adjusted HR 0.77, 95% CI 0.69–0.86].

The authors used an active comparator design and meticulously adjusted for over 100 variables to mitigate confounders. However, despite these thorough efforts, the results of this study should be interpreted with caution as several unmeasured confounders are likely to persist. For instance, the out-of-pocket cost associated with SGLT2i² may limit access for individuals with lower income and disadvantaged populations. These groups tend to experience worse outcomes in various cardiovascular conditions. This could have biased the treatment effect in favour of SGLT2i.

An important clue to detect residual confounding lies in the examining of the Kaplan–Meier curve for all-cause mortality. The difference in mortality was seen starting at Day 0 of drugs initiation, a pattern not seen in randomized trials of SGLT2i. Several trials of SGLT2i in patients with HF with preserved ejection fraction (HFpEF) found no mortality benefit.^3,4 The DAPA-HF trial found a significant reduction in mortality with the use of dapagliflozin in patients with HF with reduced ejection fraction (HFrEF). However, since HFrEF patients constitute only a minority of this study’s participants (predicted at ∼9%), this alone cannot account for the observed difference in mortality. Furthermore, in the DAPA-HF trial, the curves for mortality started to separate at 3 months which contrasts with the immediate separation seen in this study. This suggests that the mortality signal observed in this study may be spurious.

Another concern is the accuracy of diagnosis codes in claims data. A meta-analysis of studies of HF in claims data found that about one-fourth of patients with HF are not captured.⁵ Other diagnoses face similar accuracy challenges. Furthermore, as authors stated, Medicare data do not have ejection fraction to distinguish between HFpEF and HFrEF. To address this, authors used an algorithm to predict the type of HF which predicted that ∼91% of patients has HFpEF. This substantial imbalance between HFpEF and HFrEF calls this algorithm into question given that historical data showed that HFpEF accounts for approximately half of all HF cases.

Another issue is the loss to follow-up and the high rate of medication discontinuation which was not equal in study arms. In the as-treated analysis for all-cause mortality (supplement figure 5, b), 81% of the patients were censored at one year in the SGLT2i arm and 72% in the sitagliptin arm.

In summary, this article offers valuable real-world data on the benefits of SGLT2i in HF patients. Despite the significant methodology efforts made by authors, residual confounders persist. The data from randomized trials continue to be most accurate estimate of the treatment benefit of SGLT2i.

Declarations

Disclosure of Interest

All authors declare no disclosure of interest for this contribution.

References