Previous efforts to quantify the clinical risk for development of hyperglycemia requiring treatment in patients taking various antihypertensive regimens have been limited by the relative infrequency of this event and the limited range of antihypertensive agents used in large clinical trials [10]. For the same reasons, the risks of different agents have not been adequately compared. To address these issues, we conducted a case–control study of 11 855 New Jersey Medicaid enrollees newly started on hypoglycemic therapy.

Methods

Top Methods Results Discussion Author & Article Info References

Sources of Data

The study sample was drawn from the state of New Jersey's Medicaid program for the years 1981 to 1990. Enrollment in the program was ascertained through the Medicaid eligibility file, which identifies all program participants, their dates of coverage, and demographic characteristics including age, gender, and race. Data on medication use were taken from the Medicaid pharmacy claims file, which contains information on all prescriptions filled by Medicaid recipients, including the National Drug Code and the date dispensed. Medications of interest were identified using the National Drug Code specific for each agent. Hospitalization data for all Medicaid recipients were also ascertained. Medicaid recipients who were eligible to receive Medicare benefits ("crossover" enrollees) were identified to acquire full information on utilization of hospital services. Medicaid covers payment for medications for these Medicare recipients, which enabled us to ascertain all prescription information for this group.

Case Patients

Case patients were Medicaid enrollees 35 to 99 years of age who filled a first prescription for a hypoglycemic agent between 1981 and 1990. All oral hypoglycemic medications and insulin preparations were included. The date of the first prescription for a hypoglycemic medication was defined as the index date. To ensure that study patients were active users of the Medicaid system, each case patient was required to be continuously eligible in the program for at least 120 days before the index date and to have filled a prescription for at least one drug of any kind during this period. No claims for a hypoglycemic agent could be made before the index date. Drug claims were searched back to 1981 to ensure that case patients and controls had not previously filled a prescription for a hypoglycemic agent.

Of the 35 202 patients who filled a first prescription for a hypoglycemic agent during the defined study period, 443 had deficient eligibility records, and 4753 did not meet the specified age criteria. The additional requirement that case patients be continuously enrolled in Medicaid during the 120 days before the index date eliminated 9292 persons. Another 8859 patients were excluded from the case group because they had not filled a prescription for any other drug during the 120 days preceding the index date. The final case group included 11 855 patients.

Controls

We identified controls by selecting a random sample of Medicaid enrollees who met the same criteria for eligibility and use of the health care system as did case patients but who had not filled a prescription for a hypoglycemic agent on or before a randomly assigned index date. One control was randomly selected for each case.

Definitions of Antihypertensive Drug Use

To characterize current exposure to antihypertensive agents, all prescriptions filled during the 120 days before the index date were reviewed for both case patients and controls. Medications of principal interest included thiazide diuretics; angiotensin-converting enzyme inhibitors; centrally acting antiadrenergic agents; peripherally acting antiadrenergic agents; ß-blockers; calcium-channel blockers; and vasodilators. Categorization of patients taking multiple-agent regimens representing more than one category of antihypertensive agent was based on whether the regimen did or did not contain a thiazide diuretic. Drug-exposure categories were mutually exclusive so that patients were either characterized as exposed to a single category of antihypertensive therapy or placed in one of the multiple-category groups, but not both. Three time windows of exposure before the index date were created: 1 to 45 days, 1 to 90 days, and 91 to 120 days. A patient was considered to be currently exposed to an agent if he or she filled a prescription for a medication of interest in the 45 days before the index date. This was the exposure window of primary interest. In addition, we examined records of patients who filled a prescription for a medication of interest 1 to 90 days before the index date and 91 to 120 days before the index date to study people with evidence of a more prolonged duration of exposure.

Covariates

Because other medications can also directly and indirectly affect glucose metabolism, we measured exposure to potassium supplements, potassium-sparing diuretics, oral glucocorticoids, estrogen-containing compounds, and niacin. In addition, exposure status for these agents was characterized according to the previously defined exposure windows. Other covariates included demographic variables (age, gender, race, residence in a nursing home) the number of days hospitalized, and the number of prescriptions filled in the 120-day period before the index date.

Statistical Analysis

Differences in categorical variables between case patients and controls were assessed with chi-square tests. Relative risks for new use of a hypoglycemic agent within each exposure category were estimated from odds ratios calculated by unconditional logistic regression [11] using the SAS CATMOD program [12]. For these analyses, the reference category was the absence of exposure to any antihypertensive medication. Models were constructed with patients characterized according to the study definitions of duration of exposure (current exposure or exposure of more prolonged duration). Covariates in the models included age, gender, race, nursing home residency, number of days in the hospital, number of drug claims, glucocorticoid exposure, potassium supplementation, estrogen exposure, and niacin exposure. Because of temporal trends in the pattern of antihypertensive drug prescribing [13], the study sample was stratified according to index date (pre-1985 or 1985 to 1990), and models were constructed for each subsample. Confidence intervals (CIs) for the estimated odds ratios and significance tests for differences between individual odds ratios were calculated using the estimated standard errors.

To control for potential surveillance bias (that is, the possibility that patients receiving antihypertensive agents may have had closer medical supervision and may have had more laboratory tests than patients not receiving antihypertensive therapy) and to make the case and control groups more comparable, we did a separate analysis of the 8005 current users of antihypertensive medications (4794 case patients and 3211 controls). In these models, current exposure to thiazide diuretics was the reference exposure. This model also included the same potential confounding variables mentioned above.

Results

Top Methods Results Discussion Author & Article Info References

We found no consistent trend when examining the relation between age and the likelihood of starting a hypoglycemic medication Table 1 in the age range studied. Patients 55 to 74 years of age were more likely to be started on a hypoglycemic agent than patients 35 to 54 years of age, although no significant difference was found between patients 75 years of age or older and patients 35 to 54 years of age. No significant differences in gender distribution were observed between case patients and controls. Blacks and "other" non-whites were more likely to be started on hypoglycemic therapy than whites. Patients in nursing homes were less likely to start hypoglycemic therapy than noninstitutionalized patients. Patients who had a greater number of hospital days or who filled more prescriptions were more likely to be started on a hypoglycemic medication. No relation was found between index date (pre-1985 or 1985 to 1990) and the likelihood of being started on hypoglycemic therapy.

The relative risk for the initiation of hypoglycemic therapy was significantly increased among patients who were current users of antihypertensive therapy compared with nonusers; this was true for all categories of antihypertensive therapy (Table 2). When adjustments were made for demographic and health-services-utilization variables in a multivariate model Figure 1, the risk remained increased for virtually all antihypertensive exposures. The risk among current users of thiazide diuretics was 1.40 (95% CI, 1.26 to 1.58). The risk was greatest among those taking multiple-agent regimens. The relative risk for patients taking multiple-drug regimens that included a thiazide diuretic was 1.93 (CI, 1.75 to 2.13), and the relative risk among patients on multiple-agent regimens that did not include a thiazide was 1.76 (CI, 1.49 to 2.07). When the study patients were stratified according to the index date (pre-1985 or 1985 to 1990), the results were similar for each time period and did not differ from the results of the analysis for the overall sample. When case patients and controls were characterized according to the study definitions of duration of drug exposure, the results remained essentially unchanged.

View larger version (14K): [in this window] [in a new window]   Figure 1. The risk for development of hyperglycemia requiring treatment in users of antihypertensive drugs relative to nonusers. Circles indicate point estimates (odds ratios [ORs]). Horizontal bars indicate 95% CIs. Odds ratio for nonusers = 1.0. ACE = angiotensin-converting enzyme (inhibitors).

When the analysis was restricted to only case patients and controls who were current users of antihypertensive agents (n = 8005), with users of thiazide diuretics serving as the reference group (odds ratio, 1.0), no individual antihypertensive agent differed significantly from thiazide diuretics regarding the risk for initiation of hypoglycemic therapy (Figure 2). Patients receiving multiple-agent regimens remained at a significantly increased risk for being started on a hypoglycemic agent when compared with patients receiving thiazide diuretics alone.

View larger version (19K): [in this window] [in a new window]   Figure 2. Risk associated with receiving hypoglycemic therapy among users of antihypertensive drugs relative to users of thiazides. Circles indicate point estimates (odds ratios [ORs]). Horizontal bars indicate 95% CIs. The odds ratio for thiazide = 1.0. ACE = angiotensin-converting enzyme (inhibitors).

Discussion

Top Methods Results Discussion Author & Article Info References

Although the antagonistic effects of thiazide diuretics on glucose metabolism have been clearly shown, the relation of such therapy to the development of hyperglycemia requiring therapy in actual clinical practice has not been adequately examined. Previous attempts to evaluate this question have been limited to a few prospective observational studies [14, 15] and to traditional controlled clinical trials of antihypertensive therapy. For example, in a 12-year follow-up study of 226 hypertensive women (1968 to 1981), Bengtsson and coworkers [16] found that 16 of 143 patients treated with a thiazide diuretic, a ß-blocker, or both developed "clinical diabetes" but that none of the untreated participants did so. However, systolic blood pressure and body weight were higher among women who received antihypertensive therapy. In the 5-year Oslo study of patients receiving treatment for mild hypertension, only 1 patient (treated with thiazide-propranolol) of 406 in the treatment group developed diabetes [17]. In the Medical Research Council trial, the incidence of "impaired glucose tolerance" leading to withdrawal of treatment was 7.5 per 1000 patient-years among bendrofluazide-treated patients compared with 1.7 among patients receiving placebo [18]. In the study by the European Working Party on Hypertension in the Elderly, a trend toward an increased rate of initiation of hypoglycemic therapy was observed in the treatment group (21.5 per 1000 patient-years compared with 14.9 per 1000 patient-years in the placebo group), but the difference was not statistically significant [19].

Our findings indicate that, in general, patients receiving medications for hypertension are more likely to be started on a hypoglycemic agent than patients not receiving such therapy. Use of a thiazide diuretic is no more likely to lead to the initiation of hypoglycemic therapy than is any other single antihypertensive agent. However, patients receiving therapy with more than one antihypertensive agent are significantly more likely to be started on hypoglycemic therapy than are patients receiving a single antihypertensive agent.

Our results are consistent with the emerging understanding of the mechanisms common to the development of hypertension, glucose intolerance, and obesity [20-23]. Hyperinsulinemia may provide the common pathogenetic link for hypertension, obesity, and diabetes mellitus. Ferrari and coworkers [24] have suggested that impairment of insulin-mediated glucose disposal with hyperinsulinemia is a primary disturbance that may even precede the development of hypertension or recognizable glucose intolerance. As we found in our study, more intensive antihypertensive regimens were associated with a higher risk for initiation of hypoglycemic therapy than use of a single antihypertensive agent. This is probably because such patients had higher blood pressures, body weights, or both, indicating that these were the more important predictors of hyperglycemia. In clinical studies, impaired glucose tolerance correlates well with both higher blood pressure and increased body mass index [25].

In our study, we used the new prescription of a hypoglycemic agent as a therapeutic proxy for the occurrence of hyperglycemia requiring treatment. Fulfillment of the case definition required a series of physician and patient actions. To be identified as a case patient, the patient had to visit a physician, and the physician had to order and interpret a laboratory test relating to glucose metabolism and then had to respond with a new prescription for an oral hypoglycemic agent or insulin. Patients who received such prescriptions then had to fill them. Differences in patterns of physician practice or patient behavior across the various categories of antihypertensive exposure could thus lead to bias in our study. For example, a physician may be more likely to monitor serum glucose levels in a patient receiving therapy with a thiazide diuretic than in a patient treated with another antihypertensive regimen, differentially increasing the likelihood of detecting hyperglycemia and initiating therapy. However, this surveillance bias would have created a higher rate of hypoglycemic therapy among users of thiazide diuretics, which we did not find. Alternatively, physicians may see clinically meaningful hyperglycemia more commonly with thiazide diuretics than with other antihypertensive agents and may address the problem with approaches other than therapy with hypoglycemic drugs (for example, reduction in thiazide dose, correction of thiazide-induced hypokalemia, or change in antihypertensive regimen). This would result in an underestimation of the risk associated with thiazide use.

Selection factors may have also played a role in these results, potentially obscuring the true effect of thiazide diuretics relative to other agents. Thiazides may have been less frequently prescribed to more severely hypertensive patients or to patients with preexisting glucose intolerance or a family history of diabetes. Thus, our findings may reflect a channeling effect whereby patients with multiple risk factors for diabetes mellitus are more likely to be treated with nonthiazide antihypertensive medications. However, the consistent findings across the two time strata (pre-1985 and 1985 to 1990) tend to challenge this explanation. Although the effects of thiazide diuretics on glucose tolerance have been recognized since the 1960s, the increased level of awareness surrounding this issue and the widespread use of alternatives to thiazide diuretics for treatment of mild hypertension are relatively recent phenomena [26]. Even among patients with existing diabetes, physicians have not been reluctant to prescribe thiazide diuretics. The 1989 National Ambulatory Medical Care Survey indicated that for patients with a principal diagnosis of diabetes, hydrochlorothiazide was the third most frequently used medication after insulin and glyburide [27].

Our approach to this study allowed us to overcome two major limitations of previous studies: 1) the relatively small number of outcomes in longitudinal observational studies and clinical trials despite large numbers of participants and 2) the inability to examine more than a few antihypertensive therapies in a single clinical study. The use of comprehensive health claims data for a large patient sample enabled us to efficiently identify a large number of case patients meeting explicit criteria for study inclusion. The size of the study sample also made it possible to examine the full range of available antihypertensive regimens. However, our reliance on claims-based prescription data did not allow the assessment of several relevant factors, including blood pressure, body mass index, fat distribution, pre-existing glucose intolerance, family history, and physical activity level.

As recently described by Alderman [10], the growth of therapeutic options in the management of hypertension has created a practical dilemma for clinicians. The results of numerous clinical trials in which thiazide diuretics were the primary therapy must be balanced against concerns about adverse effects when making the therapeutic choice for the patient. Although our study does not suggest any advantage of one antihypertensive over any other regarding the risk for initiation of hypoglycemic therapy, the evidence does not suggest that any excess risk is associated with thiazide diuretics relative to alternative antihypertensive agents. Further research is required to confirm this finding in other large patient samples and to determine whether appropriate physician selection and management of patients taking thiazides is responsible for our finding or whether the hyperglycemic effect of these drugs may be of less clinical relevance than has been suspected.