A Systematic Review of Drug Therapy to Delay or Prevent Type 2 Diabetes

RESEARCH DESIGN AND METHODS

Detailed search strategies were designed to detect randomized controlled trials (RCTs) and cohort studies examining the effects of drug therapy on the subsequent incidence of type 2 diabetes. We searched the Cochrane Controlled Trials Registry (first quarter, 2004), MEDLINE (1966 to June, week 3, 2004), and EMBASE (1980 to week 26, 2004). Reference lists of original studies and narrative reviews were also searched manually. The search was not limited by language and is considered up-to-date as of 1 June 2004.

Studies were included if they reported, or provided sufficient data to calculate, type 2 diabetes incidence using an intention-to-treat analysis. In studies with multiple interventions, only the results of drug intervention arms compared with a placebo or control group were included.

A medical librarian (J.V.) performed the initial search with input from the other authors. The search was limited to adult patients (aged >18 years) with a minimum study sample size of 50 patients. In addition to a general drug search, a specific search for the following agents was performed: sulfonylureas, metformin, phenformin, acarbose, thiazolidinediones, insulin, hydroxymethylglutaryl (HMG)-CoA reductase inhibitors (hereafter, referred to as “statins”), estrogen, phentermine, orlistat, sibutramine, ACE inhibitors, angiotensin receptor blockers, β-blockers, calcium channel blockers, diuretics, and α-blockers. Articles were excluded if the intervention was tested in patients with preexisting diabetes, the sample size was <50 patients, the citation was a review or duplicate article, and if the study was published only in abstract form. A detailed systematic review of antihypertensive drugs and type 2 diabetes incidence (current through August 2003) has recently been published, and this review was updated by including studies published after August 2003 (11).

Two reviewers (R.P. and S.R.M.) independently examined abstracts of the remaining studies for potential inclusion and performed data extraction. Cohen’s κ coefficients were calculated to assess interobserver agreement for study inclusion and data extraction. Disagreements were resolved by consensus.

A priori, we decided that quantitative meta-analyses of data would not be possible due to substantial between-study differences in end point definitions, patient populations, and interventions.

RESULTS

Of the 5,511 initial citations, 5,222 were potentially relevant upon initial screening (online appendix [available at http://care.diabetesjournals.org]). Of these, 4,247 citations involved prevalent cases of type 2 diabetes and were excluded. After screening the titles and abstracts of the remaining 975 citations, 36 full-text articles were reviewed and 10 articles met inclusion criteria. An additional 15 articles were identified through manual searches and review of the reference lists of all included reports.

Interobserver agreement was 1.0 for study inclusion and 0.91 for data extraction.

Oral hypoglycemic agents

Ten studies, including eight RCTs, examined the effect of oral hypoglycemic agents on diabetes incidence (Table 1).

Biguanides.

The largest and most methodologically rigorous trial was the Diabetes Prevention Program (DPP), which randomized 2,155 individuals with IGT to metformin or placebo (4). After a mean follow-up period of 2.8 years, the incidence of diabetes was 7.8% in the placebo-treated patients versus 4.8% in those treated with metformin (relative risk [RR] 0.69, 95% CI 0.57–0.83); metformin was also associated with a 2.0-kg (95% CI 0.8–3.2) weight reduction compared with placebo. In post hoc subgroup analyses, the benefits of metformin were primarily observed in patients <60 years of age (RR 0.66, 95% CI 0.40–0.79 for patients 25–44 years old) and patients with a BMI ≥35 kg/m² (0.47, 0.35–0.63).

After metformin was discontinued at the end of the DPP study, patients were observed for a 1- to 2-week washout period, and the number of new cases of diabetes was ascertained (12). In the 79% of eligible patients who completed a washout visit, the incidence of diabetes increased from 25.2 to 30.6% in the metformin group and from 33.4 to 36.7% in the placebo group. When results of the washout period were included in the overall analysis, metformin still significantly decreased diabetes incidence (RR 0.75, 95% CI 0.62–0.92).

The remaining biguanide studies found no significant reduction in the incidence of diabetes compared with placebo using intention-to-treat analyses (13–17). All of these studies had very low diabetes incidence rates and were likely underpowered.

Acarbose.

Acarbose was studied in one RCT and one cohort study (13,17). In the Study To Prevent Noninsulin-Dependent Diabetes Mellitus (STOP-NIDDM) trial, the incidence of diabetes was 32% in the acarbose group and 42% in the placebo group (RR 0.75, 95% CI 0.63–0.90) during 39 months of observation (17). Nearly 25% of individuals discontinued therapy early, predominantly due to acarbose-induced gastrointestinal toxicity. At study end, 60% of eligible patients were observed for a 3-month washout period, during which 15% of acarbose-treated patients developed diabetes compared with 10.5% of placebo-treated patients. In a secondary analysis, acarbose reduced cardiovascular events from 4.7 to 2.1% (hazard ratio [HR] 0.51, 95% CI 0.28–0.95) (18).

Sulfonylureas.

Two studies examined the effect of tolbutamide therapy on diabetes incidence in patients with IGT or high-normal/elevated fasting glucose levels (19,20). Neither study reported a statistically significant decrease in the type 2 diabetes incidence compared with control or placebo, although both studies were small and potentially underpowered (Table 1).

Thiazolidinediones.

While the Troglitazone Prevention of Diabetes (TRIPOD) study reported a reduction in the incidence of type 2 diabetes from 45 to 20% (RR 0.45, 95% CI 0.25–0.83) with troglitazone (associated with a nonsignificant weight gain compared with placebo of 0.3 kg [95% CI 0.8–1.4]), the nearly 33% attrition rate during follow-up is a major limitation (21). Eight months postdrug discontinuation, type 2 diabetes incidence was assessed in approximately one-half of eligible patients, with one patient (2%) in the troglitazone arm and six patients (15%) in the placebo group developing diabetes.

One additional small cohort study found a significant reduction in diabetes incidence with thiazolidinedione therapy (Table 1) (22).

Antiobesity agents

While orlistat reduced the incidence of type 2 diabetes from 9 to 6% (RR 0.63, 95% CI 0.46–0.86) and weight by 2.8 kg (95% CI 1.1–4.5) compared with placebo in the Xenical in the Prevention of Diabetes in Obese Subjects (XENDOS) study, the attrition rate was 57% (Table 2) (23). Ninety-one percent of orlistat-treated patients experienced gastrointestinal side effects in the first year of therapy compared with 65% of the placebo arm.

A pooled analysis of three RCTs enrolling 642 obese patients reported a nonsignificant reduction in the incidence of type 2 diabetes from 2 to 0.6% with orlistat therapy (RR 0.25, 95% CI 0.05–1.2) (24). The CIs were wide, reflecting the low absolute incidence of diabetes within these trials, and attrition rates averaged >30%.

Antihypertensive drugs

A recently published systematic review of 24 studies involving antihypertensive drugs found that diabetes incidence is unchanged or increased by thiazide diuretics and β-blockers and unchanged or decreased by ACE inhibitors, calcium channel blockers, and angiotensin receptor blockers (11). Six placebo-controlled trials were included in this review. Thiazide diuretic-based treatment regimens were associated with non-statistically significant increases in the incidence of type 2 diabetes from 7.5 to 8.6% in the Systolic Hypertension in the Elderly Program (SHEP) trial (RR 1.2, 95% CI 0.9–1.5) and from 4.7 to7% in the European Working Party on High Blood Pressure in the Elderly (EWHPE) trial (1.5, 0.85–1.6) (25,26). ACE inhibitor therapy lowered diabetes incidence in the Heart Outcomes Prevention Evaluation (HOPE) trial from 5.4 to 3.6% (0.66, 0.51–0.85) and from 22 to 6% in a small group of patients with heart failure (0.26, 0.13–0.53) (27,28). Angiotensin receptor blocker therapy significantly decreased diabetes incidence in the Candesartan in Heart Failure Assessment of Reduction in Mortality and Morbidity (CHARM) study from 7 to 6% (0.78, 0.64–0.96) and resulted in a nonsignificant decrease in diabetes incidence from 5.3 to 4.3% in the Study of Cognition and Prognosis in the elderly (SCOPE) trial (0.81, 0.62–1.06) (29,30).

Overall, diabetes incidence was not a prespecified, primary end point in any study, and there was insufficient evidence to definitively recommend any given antihypertensive drug class in patients at risk of developing type 2 diabetes (11).

We found two additional studies published after the aforementioned systematic review (Table 2). In 16,176 hypertensive patients with coronary artery disease enrolled in the International Verapamil-Trandolapril Study (INVEST), a verapamil-based treatment regimen was associated with a decrease in the incidence of type 2 diabetes from 8.2 to 7% compared with an atenolol-based regimen (RR 0.85, 95% CI 0.77–0.95) (Table 2) (31). Diabetes incidence was not a predefined end point in this study and no adjustment was made for concomitant therapies, which could potentially affect diabetes incidence. In the Valsartan Antihypertensive Long-term Use Evaluation (VALUE) trial of 10,419 hypertensive patients at high cardiovascular risk, a valsartan-based treatment regimen was associated with a decrease in the incidence of type 2 diabetes from 16 to 13% compared with an amlodipine-based regimen (odds ratio [OR] 0.77, 95% CI 0.69–0.86) (32).

Statins

Four post hoc analyses of placebo-controlled statin trials reported conflicting results regarding the effect of statin therapy on diabetes incidence (Table 2). In the West of Scotland Coronary Prevention Study (WOSCOPS), diabetes incidence was significantly lower with pravastatin treatment (RR 0.70, 95% CI 0.50–0.99) (33). In the Heart Protection Study, 4.6% of simvastatin-treated patients developed diabetes versus 4.0% in the placebo arm (1.15, 0.99–1.34) and in the Long-term Intervention with Pravastatin in Ischemic Disease (LIPID) study, 4.0% of pravastatin-treated patients developed new diabetes versus 4.5% in the placebo group (0.89, 0.70–1.13) (34,35). In the Anglo-Scandinavian Cardiac Outcomes Trial-Lipid Lowering Arm (ASCOT-LLA), the incidence of diabetes was 3.0% in the atorvastatin arm and 2.6% in the placebo group (1.15, 0.91–1.44) (36).

Fibrates

In a post hoc analysis of 303 patients with IGT from the Bezafibrate Infarction Prevention (BIP) trial, bezafibrate therapy was associated with a reduction in diabetes incidence from 54 to 42% compared with placebo (HR 0.70, 95% CI 0.49–0.99) (Table 2) (37).

Estrogen

One RCT and five cohort studies have examined the association between estrogen use and diabetes incidence (Table 2). Post hoc analysis of the Heart Estrogen/Progestin Replacement Study (HERS) study reported that combination estrogen and progesterone therapy was associated with a significant reduction in the incidence of diabetes from 9.5 to 6.2% compared with placebo (RR 0.65, 95% CI 0.48–0.89) (Table 2) (38).

The Nurses Health Study, which was the largest of the cohort studies, found that over 12 years, current estrogen use was associated with a significant reduction in diabetes incidence compared with never users (RR 0.82, 95% CI 0.70–0.96) (39). Diabetes incidence in former estrogen users was not significantly different from never users (1.07, 0.95–1.2). Of the remaining four cohort studies (40–43), only one reported a significant covariate-adjusted reduction in diabetes incidence in users of estrogen replacement compared with nonusers. Several trials failed to adjust for potentially important covariates such as family history, weight, or baseline glucose measurements.

CONCLUSIONS

In summary, a number of studies have examined the impact of different drugs on diabetes incidence, including oral hypoglycemic agents, antiobesity drugs, statins, fibrates, estrogen, and antihypertensive drugs. Oral hypoglycemic medications and orlistat are the only drugs that have been studied in RCTs with diabetes incidence as the primary end point. The adequately powered studies have shown significant decreases in diabetes incidence with metformin, acarbose, troglitazone, and orlistat; however, high attrition rates were found in trials of the latter two agents. Evidence for statins, estrogen, and antihypertensive agents is conflicting and is limited to cohort studies and secondary post hoc analyses of RCTs.

A potential limitation of any systematic review (including ours) is the possibility of publication bias. In addition, studies reporting diabetes incidence as a secondary or post hoc end point were difficult to identify using standard search strategies because this information was contained within the text of studies and identifiable only by performing manual searches. Indeed, our use of manual searching and examination of bibliographies yielded more valid studies for inclusion than our original search strategy. Regardless, while the possibility of missing trials reporting secondary or post hoc analyses exists, we feel that it is unlikely that any definitive studies were missed.

Besides the reduction in glucose levels achieved by oral hypoglycemic agents, it is likely that drug-induced weight loss is contributing to the observed reduction in diabetes incidence. All but one (thiazolidinediones) of the agents reported to lower type 2 diabetes incidence directly or to indirectly promote weight loss. Weight loss has also been the target of lifestyle modification interventions in the diabetes prevention trials (9,10). In contrast to drug therapy, intensive lifestyle interventions have produced reductions in diabetes incidence of 42–58% in the three largest studies to date, despite modest degrees of weight loss of ∼5 kg or less compared with control populations (3,4,44). In the DPP, the incidence of type 2 diabetes was 3% lower in the lifestyle arm compared with the metformin arm (RR 0.61, 95% CI 0.49–0.76) and lifestyle modification was efficacious regardless of age, sex, BMI, or ethnic background (4). Assuming that such intensive lifestyle interventions can be successfully implemented in a more practical and equally effective form outside of a clinical trial setting, recidivism remains a major problem. Even within the DPP, the number of participants achieving weight loss targets (7% of initial body weight) decreased from 50% at 24 weeks to 34% at the end of follow-up, and the number of individuals who met the target exercise levels (150 min per week) declined from 74 to 58% by the end of the trial.

A critical and unresolved issue is whether drug therapy simply delays or masks the diagnosis of type 2 diabetes, rather than exerting an actual preventative effect. Drugs that acutely lower serum glucose levels may simply lower glucose concentrations to a lower cutoff level than that required for the formal diagnosis of diabetes. In the posttrial washout periods of the STOP-NIDDM and DPP trials, the higher incidence of diabetes in the treatment arms suggests that at least some of the observed benefits were merely due to delay or masking of diabetes. It is unknown if the beneficial effects of the drugs would have persisted if the posttrial follow-up periods were longer. In the TRIPOD study, diabetes incidence, β-cell function, and insulin sensitivity remained stable in the troglitazone arm for at least 8 months after drug discontinuation (21). However, the type 2 diabetes incidence in the posttrial period was very low, and reasons for the high attrition rates in the study were not detailed.

To prove that diabetes is actually prevented, future studies will have to demonstrate arrest of the disease process. Because the average time interval between onset of β-cell dysfunction and development of type 2 diabetes is 10 years (45), follow-up periods will have to be substantially longer. In a recent retrospective cohort analysis of 10,996 patients with diabetes newly treated with oral agents, statin therapy was associated with a 10-month delay in the initiation of insulin therapy (46). However, after 7 years of follow-up, there were no differences between the statin and control groups in their requirements for insulin therapy.

It will also be important to demonstrate a reduction in morbidity and mortality in order to accept that any of these drugs are beneficial in patients who have not yet developed diabetes. The finding that acarbose reduced cardiovascular events in the STOP-NIDDM trial was based on a small number of events and will require confirmation.

There are a number of potential adverse effects associated with drug-related diabetes prevention strategies. For example, hypoglycemia is a potentially limiting side effect of sulfonylurea therapy, occurring at a frequency of 3% in patients with IGT enrolled in the Fasting Hyperglycemia Study (47). Gastrointestinal toxicity contributed to high discontinuation rates in the STOP-NIDDM and XENDOS studies, and troglitazone is no longer available because of the risk of serious hepatotoxicity (48). Given the likelihood of long-term therapy with diabetes prevention agents, additional data regarding adverse events and adherence will be required.

In addition to their clinical effectiveness in diabetes prevention, consideration should also be given to the cost-effectiveness of drug interventions. Two economic analyses of the DPP study have been performed (49,50). In a cost-effectiveness analysis from a societal perspective, the metformin intervention cost $31,300 per case of diabetes delayed or prevented and $99,600 per quality-adjusted life year gained over the 3-year duration of the study (49). Assuming the use of lower-priced generic metformin, cost estimates decreased to $14,300 and $35,000, respectively. In all analyses, the lifestyle intervention was more economically attractive than metformin. A second economic analysis, performed in Europe, also factored in estimates of cost savings for each case of diabetes presumably prevented (50). Metformin was found to be cost-saving in four of the five European countries studied.

A number of currently ongoing studies should provide more definitive evidence (Table 3) (47,51–56). Short-acting insulin secretagogues, renin-angiotensin inhibitors, newer thiazolidinediones, and insulin glargine are among the drug classes being investigated. The majority of these studies are scheduled for completion in the latter half of this decade.

In conclusion, a number of studies have investigated the effects of several different drug classes on type 2 diabetes incidence. The available evidence suggests that oral hypoglycemic drugs may reduce diabetes incidence compared with placebo, while the evidence for orlistat, statins, estrogen, and antihypertensive drugs is inconclusive. However, the data are not definitive and no single agent can currently be recommended for diabetes prevention. It is critical that future studies be designed with much longer follow-up periods and with development of new-onset diabetes as the primary outcome, so as to differentiate between genuine diabetes prevention as opposed to simple delay or masking of this condition.