Thiazolidinediones and Heart Failure: A teleo-analysis

Data from RCTs, observational studies, and anecdotal case reports were analyzed in a teleo-analysis. A teleo-analysis attempts to determine the adverse effect of a drug by complementing information from different study designs across all grades of evidence (12). Controlled data from randomized trials and observational studies are used to evaluate the statistical association between a drug and its adverse effect and estimate the frequency of the adverse event. This is complemented by data from individual case reports, allowing for the evaluation of specific characteristics of adverse effects such as dose, time, and susceptibility factors. We searched without language restrictions and contacted authors when specific aspects of the data required clarification.

We retrieved potentially relevant citations by examinging trials identified in existing published meta-analyses (13,14). PubMed was searched from January 2003 to September 2006 for recent trials by applying the “Randomized Controlled Trial” search filter to the medical subject heading terms “pioglitazone” or “rosiglitazone,” which yielded 221 citations. We retrieved unpublished RCTs for rosiglitazone from the manufacturer's Web site (15). No unpublished RCTs were available for pioglitazone. Included RCTs 1) were at least 6 months’ duration of TZD for treating or preventing type 2 diabetes, 2) were direct head-to-head evaluations between a TZD alone (rosiglitazone or pioglitazone) and placebo alone, and 3) provided numerical data on patients experiencing heart failure.

PubMed was searched using the medical subject heading terms “thiazolidinediones“ and “case-control studies” or “cohort studies” in September 2006. We used the bibliographies of relevant studies, the “Web of Knowledge Cited References” list, and the “Related Articles” link in PubMed to identify additional studies. Included articles were controlled observational studies containing data that enabled us to calculate the OR of new onset of heart failure developing in patients receiving TZDs compared with patients receiving other oral antidiabetes medications with or without insulin.

PubMed, Embase, and Google Scholar were searched using the terms “thiazolidinediones” and “heart failure” to identify case reports and case series in September 2006. We used the bibliographies of relevant studies, the “Web of Knowledge Cited References” list, and the “Related Articles” link in PubMed to identify additional studies. We identified 28 published reports. All adverse drug reactions reported with pioglitazone (n = 195) or rosiglitazone (n = 830) to Health Canada's spontaneous adverse event reporting system (CADRMP), the only English language regulatory authority database that allows easy and unrestricted Web access, were reviewed in September 2006 to identify reports of heart failure or pulmonary edema (16). Reports describing only edema were excluded. A total of 51 reports of heart failure with pioglitazone and 163 reports of heart failure with rosiglitazone were identified. Information on age, sex, duration, dose of TZD, outcome of reaction, and concomitant medications was abstracted from both the published case reports and the spontaneous reports.

We used RevMan 4.28 to calculate summary ORs (random-effects model) and heterogeneity.

We identified three trials of 10,731 patients that provided numerical information on heart failure events (Table 1) (10,17,18). The pooled OR for heart failure was 2.1 (95% CI 1.08–4.08; P = 0.03) (Fig. 1). There is moderate heterogeneity around this estimate (I² = 59%), which is attributable to the different populations and varying criteria for ascertaining heart failure events.

Four observational studies of 67,382 patients were included (19–22). The pooled OR for heart failure with TZDs was 1.55 (95% CI 1.33–1.80; P < 0.00001) (Fig. 1). Moderate heterogeneity was found (I² = 47%) due to the differences in population and coexisting morbidity. If we exclude the study by Inzucchi et al. (21), which focused on diabetic patients after myocardial infarction rather than on diabetic patients in general, sensitivity analysis shows the heterogeneity to have an I² of 0%.

Two observational studies were excluded because they did not fulfill the selection criteria (Table 2) (23,24). The study by Masoudi et al. (23) studied readmission rates for patients on TZDs who already had a recent admission for heart failure, whereas the study by Rajagopalan et al. (24) compared the rates of heart failure solely between patients on pioglitazone and insulin. Our selection criteria were based on patients potentially being able to receive other oral hypoglycemic agents, with or without insulin. The use of a control group comprising patients on insulin is particularly prone to confounding, as patients with the most severe type 2 diabetes are likely to receive insulin.

A dose-time-susceptibility analysis of 162 analyzable case subjects showed that the age ranged from 31 to 88 years, with a median age of 67 years. The adverse effect was not limited to the elderly because 42 of 162 case subjects (26%) were aged <60 years.

Among 99 analyzable cases, the median duration for the onset of heart failure was 24 weeks (range 1–260), suggesting that the reaction is delayed or late onset in nature (Fig. 2). Patients on low doses were also at risk of developing heart failure soon after starting treatment. One patient on pioglitazone and nine patients on rosiglitazone developed heart failure in the first 4 weeks of low-dose therapy (defined as ≤15 mg pioglitazone or ≤4 mg rosiglitazone). Heart failure did not develop much earlier in those receiving TZDs in the high-dose range (Fig. 3). Heart failure due to TZD use also did not occur as a function of the cumulative ingested dose. Heart failure can occur even at doses below those used for therapeutic benefit, and the relatively low daily doses of 15 mg pioglitazone or 2 mg rosiglitazone are already at the top end of the dose-effect curve for triggering episodes of heart failure.

We were unable to identify any particular susceptibility factors, and heart failure occurred equally among 110 men and 88 women. Death occurred in nine patients.

Our teleo-analysis provides useful clinical information on the increased risk of heart failure with TZDs. This evidence should be judged in the context of a recent systematic review of pioglitazone, which found little evidence for beneficial effects on hard clinical end points (25).

Diabetes and the use of any therapy for the treatment of diabetes is associated with an increased risk of heart failure (26). The background incidence of heart failure in patients with diabetes is ∼1.9% over 2.2 years (27). The estimated number needed to harm with TZDs, based on an OR of 2.10, would be ∼50 over a 2.2 year follow-up period.

The development of heart failure is a class effect of the TZDs, mediated through increased plasma volume rather than through a direct effect on the myocardium (9,17,28). A recent meta-analysis of RCTs found the OR for edema with TZDs to be 2.26 (95% CI 2.02–2.53) (14), similar to our estimates of heart failure. Fluid retention due to TZDs may trigger (clinically apparent) episodes of heart failure in susceptible individuals or may unveil the disease in those with latent heart failure (no previous cardiac history). This theory is supported by echocardiographic evaluation of heart failure patients where no deterioration in left ventricular function was found with TZDs therapy (5,17). Similarly, a 52-week study of rosiglitazone did not show any decline in the left ventricular ejection fraction (29). Fluid retention appears to be mediated through increased sodium reabsorption by the renal peroxisome proliferator–activated receptor-γ–dependent pathway in the collecting tubules, suggesting a possible therapeutic role for amiloride rather than loop diuretics, for which resistance has been reported (30).

The rosiglitazone package insert warns against the use of rosiglitazone in New York Heart Association (NYHA) Stage III and IV heart failure patients and cautions about the increased risk of heart failure in combination with insulin (31). Similarly, the package insert of pioglitAzone cautions against this increased risk of heart failure (32). However, this increased risk is not confined to patients on insulin. None of the patients in the DREAM (Diabetes REduction Assessment with ramipril and rosiglitazone Medication) Trial were on insulin (10), and only one-third of patients required insulin in the PROActive (PROspective pioglitAzone Clinical Trial In macroVascular Events) Study (18). We were unable to obtain information on concomitant medications in GlaxoSmithKline Study 211 (17) (Table 1).

A 2003 consensus review by the American Heart Association and the American Diabetes Association rightly recommends that TZDs be avoided in patients with NYHA Stage III and IV heart failure (as RCTs of TZDs excluded these patients) (9). However, they also recommend that TZD therapy be initiated at a low dose, slowly increased, and closely monitored in diabetic patients with NYHA Stage I and II heart failure and in those with depressed ejection fractions (<40%) and one or more cardiac risk factors without heart disease (9). Recommendations for their use, even at low doses in NYHA Stage 1 and II heart failure patients and in those with risk factors for heart failure, may need to be carefully revaluated as heart failure occurred even among patients with no history of heart failure in the DREAM Trial and observational studies. Heart failure also occurred at the lowest-dose range for TZDs in spontaneous reports and at both low (4 mg) and high (8 mg) doses of rosiglitazone used in clinical trials. The occurrence of heart failure several months after initiation of TZDs suggests a long-term effect of the drug, which may not be avoided by slow-dose titration, and mandates the need for long-term vigilance.

Further research studies should evaluate intraclass differences in the risk of heart failure between the TZDs and optimal management strategies in patients experiencing heart failure on TZDs, including immediate withdrawal and the role of diuretics such as amiloride or spironolactone (33). Studies distinguishing between heart failure requiring hospitalization versus heart failure that can be managed as an outpatient are also needed.

There are a number of limitations to our study. Diagnosing heart failure is challenging and may require subjective clinical interpretation of symptoms, signs, and additional investigations (such as chest radiographs or echocardiography). Inconsistent diagnostic criteria across studies may have led to some heterogeneity. Patients with progressive pedal edema from fluid retention could potentially have been erroneously classified as heart failure cases. This, however, seems unlikely here, since all three included RCTs were designed to actively seek out cases of heart failure and heart failure information in two of the large published studies was subjected to independently masked adjudication based on prespecified criteria (Table 1). Moreover, three of the four observational studies used fairly stringent criteria in that the patients required hospital admission due to heart failure.

In addition, anecdotal case reports may not be entirely representative of the population, were often incomplete, and were not necessarily causal, as they cannot rule out the role of concomitant medicines (metformin and insulin) in exacerbating heart failure. Case reports are subject to publication bias, which was minimized by the use of spontaneous reports from the CADRMP database. Spontaneous reports and case reports represent the observation of the individual health professionals and layperson and should not be used for numerically estimating the risk of advanced diabetic retinopathy, as the denominator is unknown. We used data from RCTs and observational studies to statistically determine the association between TZDs and heart failure. Observational studies are susceptible to confounding. We corroborated observational data with results from RCTs, and the risk of heart failure was consistent in magnitude and direction.

In conclusion, our teleo-analysis confirms the increased magnitude of the risk of heart failure with TZDs. The number needed to harm with TZDs was estimated to be ∼50 over 2.2 years. Heart failure can occur at both high and low doses, usually weeks to months after initiation of TZDs. It can occur in the absence of insulin, even in patients without a history of heart failure. Existing guidelines and package inserts may have to be revised to incorporate these risk characteristics of TZDs.