Secondary Logo

Journal Logo


Paradoxical Relations of Drug Treatment with Mortality in Older Persons

Glynn, Robert J.1,2; Knight, Eric L.1; Levin, Raisa1; Avorn, Jerry1

Author Information
  • Free


Patient characteristics including age, comorbidity, and prognosis affect prescribing decisions, even for conditions unrelated to a patient’s main medical problem. For example, among older individuals, those with emphysema are less likely to be prescribed lipid-lowering treatment, 1 and greater comorbidity decreases the likelihood that an older person will receive effective therapy for conditions including hypertension, 2 diabetes mellitus, 3,4 myocardial infarction, 5 and glaucoma. 6 Older age is itself associated with less aggressive treatment for a number of conditions, even after adjustment for comorbidity and contraindications. 4,7–9 Such reduced treatment in older patients with comorbidity may reflect inappropriate undertreatment, 1,10 uncertainty about the value of treatment arising because these patients have commonly been excluded from clinical trials, 11,12 or a calculated decision that the potential benefits of treatment are lower or the risks greater in patients with shortened life expectancy. 13 Regardless of their cause, these complex relations between comorbidity and drug use can substantially influence the apparent relation between drug use and subsequent mortality.

Comorbidity can also affect adherence to prescribed medications, as patients with greater comorbidity may have reduced adherence to therapies for unrelated conditions. 14,15 Results from both the placebo and active treatment arms of randomized trials demonstrate that adherence declines in patients near death. In both the Coronary Drug Project 16 and the Physicians’ Health Study, 17 nonadherence in both the placebo and active treatment groups was strongly associated with an increased risk of death. As patients face the burden of a serious illness, they (and their physicians) may well lose interest in preventive therapies for an unrelated condition.

From a research perspective, many studies have used data on filled prescriptions to study patterns of drug use, determinants of potential undertreatment, and the relation of drug use to adverse and beneficial effects in clinical practice. 18 Valid interpretation of such studies requires consideration of the factors associated with the decision to prescribe, the choice of treatment, and adherence to treatment. 19–21 Nonetheless, comorbidity indices based on drug use have been proposed and have some predictive validity, 22–24 although their performance in high-risk elderly populations is unclear. Our hypothesis was that selective under-use of drugs by elderly patients with substantial but unmeasured comorbidity would lead to artifactual protective associations between use of specific drugs and mortality. To address this question, we examined data on drug use and mortality in a large elderly population.

Subjects and Methods


We linked data from New Jersey Medicare’s Provider Analysis and Review file; the Medicare Denominator file, which includes date of death for all Medicare enrollees; and utilization files from New Jersey’s Medicaid and Pharmacy Assistance for the Aged and Disabled programs. Our primary study population consisted of all residents of New Jersey who, according to Medicare records, were hospitalized at least once between January 1, 1991 and December 31, 1994; were 65–99 years of age at the time of hospitalization; and satisfied minimal requirements for participation (defined below) in either the New Jersey Medicaid or that state’s Pharmacy Assistance for the Aged and Disabled program. We determined participation in these programs through review of pharmacy claims files that contain information on all prescriptions filled by enrollees, and that specify the product’s National Drug Code and the date dispensed. For all analyses, all traceable person-specific identifiers were replaced with coded study subject numbers to protect the privacy of program participants. To ensure enrollment in these programs, all study participants were required to have filled at least one prescription in either program in the 120 days before hospitalization. To document adequate time in these programs to fill all relevant prescriptions, all participants were required to have filled another prescription between 120 days and 365 days before hospitalization. For participants with more than one hospitalization during the study period, we selected one hospital stay at random for study.

The New Jersey Medicaid program has no deductible and no maximum benefit, and charges no copayment for prescription drugs, ensuring virtually complete ascertainment of drug use. New Jersey’s Pharmacy Assistance for the Aged and Disabled program also has no deductible and no maximum benefit, but there is a nominal $2 copayment for each prescription. Because its income-eligibility criteria are more generous than those of Medicaid, the Pharmacy Assistance for the Aged and Disabled program includes patients above the poverty level. During the study period, eligibility required an annual income below $15,700 if single and $19,250 if married. This high income ceiling accounts for a large and more representative population of program participants.

A total of 131,844 hospitalized individuals met the drug use criteria. We excluded 2,733 individuals with missing information on race, leaving 129,111 for analysis. Although individuals enrolled in these programs who filled no prescriptions during this period would be missed, this number is likely to be small for several reasons: about 90% of elderly individuals enrolled in prescription drug programs regularly fill prescriptions, 1 entitlement programs such as Medicaid and the Pharmacy Assistance for the Aged and Disabled program commonly enroll more impaired individuals with greater need of prescription drugs, and enrollees have a strong financial incentive to use these programs when filling prescriptions.

To examine potential selection bias associated with hospitalization, we also studied a random sample of 132,071 individuals 65–99 years of age with a randomly selected outpatient physician visit as the index event. These individuals were also required to have filled at least one prescription in the 120 days, and another between 121 and 365 days, before this outpatient visit.

Classification of Drugs

We used the specific therapeutic class coding system provided by First DataBank as an initial approach to determine the most commonly used drug classes. 25 Based on this system, the 10 most commonly used classes of drugs in the hospitalized cohort, each used by at least 20,000 subjects, were: thiazide and related diuretics, calcium channel blockers, vasodilators, anti-anxiety agents, histamine2 inhibitors, digitalis glycosides, nonsteroidal anti-inflammatory agents, angiotensin-converting enzyme inhibitors, narcotic analgesics, and potassium supplements. We reviewed the specific drugs within each class for consistency, and, on this basis, separated the thiazide and related diuretics into two separate classes: thiazide diuretics and loop diuretics.

We also selected for study nine other drug classes, each used by at least 5,000 of the hospitalized subjects. Among these, we specifically selected six other commonly used classes of cardiovascular agents so that we could compare their relative associations with mortality: beta blockers, sympatholytics, lipotropics, antiarrhythmics, hemorheologics, and oral anticoagulants. In addition, because our previous research suggested that antidiabetic drugs are less likely to be prescribed to elderly patients with comorbidity, 4 we also studied oral hypoglycemics and insulin. Finally, because we expected that comorbidity would affect use of specialty medical care, we studied glaucoma drugs.


The primary outcome variable was death occurring either in the hospital or within 1 year of discharge. We used both Medicare and Medicaid eligibility files to identify decedents 26 and had complete 1-year follow-up information on all participants.

Other Variables

We considered seven variables that might confound the relation of drug use with mortality: age, sex, race, nursing home residence, prescription program, and two measures of comorbidity. Age, sex, and race were obtained from Medicare eligibility files. We used both Medicare and Medicaid claims files to identify patients who were in a nursing home at any time in the 120 days before their hospital admission. As a measure of comorbidity and of overall drug utilization, we counted the number of different therapeutic classes in which an individual filled prescriptions during the 120 days before hospitalization or the index visit for the population sample. To quantify the severity of comorbidity in the hospitalized cohort, we used the Charlson comorbidity index, 27 calculated as recommended by Deyo et al.28 In the nonhospitalized cohort, we similarly calculated this index on the basis of outpatient diagnoses from physician visits occurring on or up to 120 days before the index date. Finally, we also classified patients according to their participation in the two prescription programs (that is, filled prescriptions in Medicaid only, in the Pharmacy Assistance for the Aged and Disabled program only, or in both programs in the 6 months before admission.)

Statistical Analysis

For each of the 20 drug classes of interest, we first calculated crude death rates, as well as age- and sex-adjusted death rates. Study time began on the date of hospitalization, or on the index date for nonhospitalized patients, and continued for 1 year after hospital discharge or the index date. We used proportional-hazards models to determine the age- and sex-adjusted relative rate of death among users of that drug class compared with all study participants who did not fill prescriptions in that drug class within 120 days of their index date. 29 To explore the potential confounding effects of other demographic variables and measures of comorbidity, we fitted additional proportional-hazards models controlling for age, sex, race, nursing home residence, prescription drug plan, Charlson comorbidity index, and number of drug classes in which prescriptions were filled. Alternative models compared users of drugs from each drug class to a common reference group consisting of those persons who met eligibility requirements but filled no prescriptions in any of the 20 drug classes of interest during the 120 days before their index date. Use of this common reference group, however, had little impact on the relative death rates, and these results are not presented.

We performed subgroup analyses to examine the potential effects of proximity to death and high levels of comorbidity on our results. Specifically, we fitted separate proportional-hazards models examining relative death rates in the first 90 days of follow-up and thereafter. Comparison of these results also provided an evaluation of the tenability of the proportional-hazards assumption. We classified a person as having a high level of comorbidity if he or she resided in a nursing home, had a Charlson comorbidity index of 3 or higher, or used drugs from 10 or more drug classes within 120 days of the index date. We then fitted separate proportional-hazards models among those who did and did not have such a marker of high risk.


More than 20% of the hospitalized cohort and the nonhospitalized sample were 85–99 years of age (Table 1). Women constituted the substantial majority of both cohorts, and many subjects resided in nursing homes (24% in the hospitalized cohort and 17% in the nonhospitalized sample).

Table 1
Table 1:
Demographic Characteristics of Hospitalized Elderly Residents of New Jersey with Drug Benefits and a Nonhospitalized Sample

Age- and Sex-Adjusted Death Rates

Overall, 41,930 of the 129,111 hospitalized individuals (32.5%) died within a year of their index hospitalization (Table 2). Users of drugs from 7 of the 20 drug classes had reduced rates of death relative to non-users. Relative to non-users of drugs from that class, those who filled prescriptions for any of the following drugs during the 120 days before their hospitalization had reduced death rates in the following year: lipid-lowering drugs, 44% reduction [95% confidence interval (95% CI) = 40–47%]; nonsteroidal anti-inflammatory agents, 26% reduction (95% CI = 24–28%); beta blockers, 24% reduction (95% CI = 22–27%); thiazide diuretics, 18% reduction (95% CI = 15–20%); glaucoma drugs, 17% reduction (95% CI = 14–20%); calcium channel blockers, 10% reduction (95% CI = 8–12%); and anti-anxiety drugs, 4% reduction (95% CI = 2–7%). In the nonhospitalized sample, we found similar effects associated with use of drugs from any of five of these categories. In this cohort, users of lipotropics, nonsteroidal anti-inflammatory drugs, beta blockers, thiazides, and glaucoma drugs had substantial reductions in death rates, ranging from 55% to 21%.

Table 2
Table 2:
Drugs Associated with Decreased Risk of Death in Either Hospitalized or Nonhospitalized Elderly Residents of New Jersey

By contrast, users of drugs from 11 of the 20 drug classes had increased age- and sex-adjusted death rates relative to non-users of these classes, ranging in the hospitalized cohort from a 5% increased death rate among users of oral hypoglycemics to a 65% increased rate among users of loop diuretics (Table 3). Use of these same drugs was also associated with increased death rates in the nonhospitalized sample. Drug classes associated with a more than 50% increase in death rate in the nonhospitalized sample were: vasodilators, narcotic analgesics, anti-arrhythmics, potassium replacements, digitalis, and loop diuretics.

Table 3
Table 3:
Drugs Associated with Increased Risk of Death in Hospitalized and Nonhospitalized Elderly Residents of New Jersey*

Rates Adjusted for Comorbidity

Control for additional demographic measures (race and prescription program) and measures of comorbidity (nursing home residence, Charlson comorbidity index, and number of drug classes in which prescriptions were filled) had little impact on the apparent “protective” association of use of certain drug classes with mortality in either the hospitalized cohort (Figure 1) or nonhospitalized sample (Figure 2). In the hospitalized cohort, users of drugs from the same seven drug classes noted above had reduced death rates ranging from an 11% reduction among users of anti-anxiety drugs and users of thiazides to a 40% reduction among users of lipotropics. In the nonhospitalized sample, effects associated with use of these drugs were all slightly larger, ranging from a 14% reduction among users of anti-anxiety drugs to a 52% reduction among users of lipotropics.

Adjusted risk of death in the hospitalized sample with 95% confidence intervals. Each relative risk was estimated from a separate proportional-hazards model comparing users to nonusers of each drug class and controlling for age, sex race, nursing home residence, prescription program, Charlson comorbidity index, and number of drug classes in which prescriptions were filled. NSAIDS = nonsteroidal anti-inflammatory drugs; ACE = angiotensin-converting enzyme.
Adjusted risk of death in the nonhospitalized sample with 95% confidence intervals. Each relative risk was estimated from a separate proportional-hazards model comparing users with nonusers of each drug class and controlling for age, sex race, nursing home residence, prescription program, Charlson comorbidity index, and number of drug classes in which prescriptions were filled. NSAIDS = nonsteroidal anti-inflammatory drugs; ACE = angiotensin-converting enzyme.

For drugs associated with increased death rates relative to non-users, however, control for these confounding variables substantially attenuated the apparent effects. After such adjustment, the relative risk of death in the hospital cohort remained elevated by more than 10% for only four drug classes compared with non-users: users of potassium replacements, anti-arrhythmics, digitalis, and loop diuretics had death rates from 19% to 38% higher than non-users of these drugs. In the nonhospitalized sample, users of drugs from these same four classes continued to have higher death rates compared with non-users, but these relative rates were also substantially reduced from the rates observed in the age- and sex-adjusted analyses.

Subgroup Analyses

We observed little modification of apparent protective effects by level of comorbidity in multivariate analyses controlling for potential confounding variables. In both the low- and high-risk strata, apparent protective effects associated with use of lipotropics, nonsteroidal anti-inflammatory agents, glaucoma drugs, beta blockers, calcium channel blockers, thiazide diuretics, and anti-anxiety drugs ranged from 12% to 53% reductions in death rates in the hospitalized and nonhospitalized cohorts. For drugs strongly associated with increased relative rates, we observed relative rates above 1 in both the high- and low-risk strata, although effects were larger among those in the low-risk strata. We also found little evidence for modification of effects for drugs associated with either increased or decreased rates of death in either cohort in analyses stratified by follow-up time (that is, early deaths vs later deaths).


Our results suggest that many effective drugs are not used by patients near death, either because of selective nonprescription by physicians or nonadherence by patients. One possible explanation is that patients with substantial comorbidity are less likely to have their unrelated disorders treated. 1 Asymptomatic conditions including elevated cholesterol, glaucoma, and hypertension are less likely to be treated vigorously in patients with the greatest risk of death. To be sure, the value of treatment for these conditions is also less clear in the oldest and frailest patients, both because they may not live long enough to benefit from these treatments, and because randomized trials have commonly not included such patients. Nevertheless, many frail older people might benefit from more aggressive drug treatment for these conditions. 7–10

Another possible explanation for our findings is that the use of some drugs may actually lower the risk of death in this population. For example, antihypertensive drugs have a documented ability to reduce mortality in elderly individuals with hypertension 30 and cholesterol-lowering drugs may have a similar effect in older patients with elevated lipid levels. 31,32 The benefits of these therapies, however, were documented in studies comparing treated individuals with untreated individuals with hypertension (or elevated cholesterol), whereas the comparison group in our study included people without these conditions who should theoretically be at lower risk of death. Indeed, levels of blood pressure in treated individuals generally remain above those of untreated persons without hypertension. 33 A true mortality benefit is even more unlikely to be the mechanism underlying other associations we observed, such as with nonsteroidal anti-inflammatory drugs, anti-anxiety drugs, and glaucoma medications. In reality, what appear to be protective effects of nonsteroidal anti-inflammatory drugs on mortality may actually reflect advice against prescribing these drugs to the oldest and sickest individuals because of their increased risk of bleeding. 34 Similarly, treatment with anti-anxiety drugs is often discouraged in frail patients who are at increased risk of falls and hip fracture. 35 Thus, the best explanation for the apparent protective effects of use of certain drug classes is that they are artifacts of selective prescribing and adherence.

We measured several potentially confounding variables with strong relations with mortality, including the Charlson comorbidity index, the number of drug classes in which prescriptions were filled, and nursing home residence. Control for these variables, however, could not explain either the apparent protective effect of use of some drug classes or the increased risk associated with use of others. Use of medications including loop diuretics, digitalis, potassium replacements, and anti-arrhythmics is independently associated with increased risk of death; such use reflects underlying diseases that are not adequately captured in the other measures of comorbidity. These findings support the value of drug indices based on some medications 23–25 to measure comorbidity and disease severity not identified by other measures of patient health. Conversely, use of medications from other drug classes, including lipotropics, nonsteroidal anti-inflammatory drugs, glaucoma drugs, beta blockers, calcium channel blockers, thiazide diuretics, and anti-anxiety drugs, was independently associated with reductions of greater than 10% in risk of death in both study samples after adjustment for potential confounding variables.

The major limitation of our study is the absence of enough detailed information from individual patients to warrant a conclusion that valuable prescriptions were withheld or unnecessary drugs were prescribed. Nonetheless, large claims datasets are commonly used to study potential under-use of medication in general populations and, as noted above, available powerful predictors of mortality could not explain the paradoxical protective effects of some medications on mortality. Indeed, controlling for these measures of comorbidity led, in some cases, to even larger “protective” associations for some drugs, such as anti-anxiety drugs and calcium channel blockers. In evaluating the potential effects of confounding by indication, absence of reduction of apparent effects upon control for confounding variables is commonly interpreted as supportive evidence that confounding is limited. 36,37 Nevertheless, the potential exists for differential errors in measuring comorbidity in users and nonusers of specific drugs. 38 Thus, because of the low likelihood that use of drugs such as glaucoma medications and nonsteroidal anti-inflammatory drugs actually lowers the risk of death, we believe that the observed protective associations are likely to be artifacts of prescribing patterns or patient adherence.

Our study population was poorer and somewhat sicker than the general elderly population, as expected because older adults with serious chronic illnesses are particularly likely to “spend down” their assets and thus qualify for drug benefit programs, and as documented by the high 1-year death rates. Thus, these results may not be completely generalizable to healthier or wealthier elderly individuals. Use of the current population does, however, remove one important source of bias associated with socioeconomic status. Although medication costs are an important and increasingly common barrier to drug treatment in older persons, enrollees in the programs studied had no limits on the number or costs of drugs that were covered. Furthermore, they had no formulary restrictions during the study period. Thus, our results were not biased because individuals had limited choices among alternative therapies or because they had to choose only those drugs that they could afford. As the debate continues about government programs to provide drug coverage for older people, our results suggest that drug utilization is affected by other factors in addition to cost.

These results also highlight the need for studies to evaluate the risks and benefits of commonly prescribed drugs in older people. Non-use of specific drugs may be a poor indicator of the absence of common diseases, and may reflect instead physician uncertainty on the actual risks and benefits of a given drug therapy in older patients. Such uncertainty is in turn often based on a dearth of clinical trial data evaluating drugs in such patients. 11 Whereas the usual expectation is that confounding by indication leads to artificially increased risks of death associated with use of drugs, selective under-use of drugs can lead to the opposite phenomenon of artificially lowered risks of death. These data provide further evidence of the perils of attributing causality to specific interventions in observational studies of clinical outcomes.


1. Redelmeier DA, Tan SH, Booth GL. The treatment of unrelated disorders in patients with chronic medical diseases. N Engl J Med 1998; 338: 1516–1520.
2. Gambassi G, Lapane K, Sgadari A, Landi F, Carbonin P, Hume, Lipsitz L, Mor V, Bernabei R. Prevalence, clinical correlates, and treatment of hypertension in elderly nursing home residents. SAGE (Systematic Assessment of Geriatric Drug Use via Epidemiology) Study Group. Arch Intern Med 1998; 158: 2377–2385.
3. Glynn RJ, Monane M, Gurwitz JH, Choodnovskiy I, Avorn J. Agreement between drug treatment data and a discharge diagnosis of diabetes mellitus in the elderly. Am J Epidemiol 1999; 149: 541–549.
4. Glynn RJ, Monane M, Gurwitz JH, Choodnovskiy I, Avorn J. Aging, comorbidity and reduced rates of drug treatment for diabetes mellitus. J Clin Epidemiol 1999; 52: 781–790.
5. McLaughlin TJ, Soumerai SB, Willison DJ, Gurwitz JH, Borbas C, Guadagnoli E, McLaughlin B, Morris N, Cheng SC, Hauptman PJ, Antman E, Casey L, Asinger R, Gobel F. Adherence to national guidelines for drug treatment of suspected acute myocardial infarction in community hospitals: evidence for undertreatment in women and the elderly. Arch Intern Med 1996; 156: 799–805.
6. Glynn RJ, Gurwitz JH, Bohn RL, Monane M, Choodnovskiy I, Avorn J. Old age and race as determinants of initiation of glaucoma therapy. Am J Epidemiol 1993; 138: 395–406.
7. Soumerai SB, McLaughlin TJ, Spiegelman D, Hertzmark E, Thibault G, Goldman L. Adverse outcomes of underuse of beta-blockers in elderly survivors of acute myocardial infarction. JAMA 1997; 277: 115–121.
8. Barakat K, Wilkinson P, Deaner A, Fluck D, Ranjadayalan K, Timmis A. How should age affect management of acute myocardial infarction? A prospective cohort study. Lancet 1999; 353: 955–959.
9. McAlister F, Taylor L, Teo KK, Tsuyuki RT, Ackman ML, Yim R, Montague TJ. The treatment and prevention of coronary heart disease in Canada: do older patients receive efficacious therapies? J Am Geriatr Soc 1999; 47: 811–818.
10. Wetle T. Age as a risk factor for inadequate treatment. JAMA 1987; 258: 516.
11. Gurwitz JH, Col NF, Avorn J. The exclusion of the elderly and women from clinical trials in acute myocardial infarction. JAMA 1992; 268: 1417–1422.
12. Bugeja G, Kumar A, Banerjee AK. Exclusion of elderly people from clinical research: a descriptive study of published reports. BMJ 1997; 315: 1059.
13. Welch HG, Alertsen PC, Nease RF, Bubolz TA, Wasson JH. Estimating treatment benefits for the elderly: the effect of competing risks. Ann Intern Med 1996; 124: 577–584.
14. Gurwitz JH, Glynn RJ, Monane M, Everitt DE, Gilden D, Smith N, Avorn J. Treatment for glaucoma: adherence by the elderly. Am J Public Health 1993; 83: 711–716.
15. Monane M, Bohn RL, Gurwitz JH, Glynn RJ, Levin R, Avorn J. The effects of initial drug choice and comorbidity on antihypertensive therapy compliance: results from a population-based study in the elderly. Am J Hypertens 1997; 10: 697–704.
16. Coronary Drug Project Research Group. Influence of adherence to treatment and response of cholesterol on mortality in the Coronary Drug Project. N Engl J Med 1980; 303: 1038–1041.
17. Glynn RJ, Buring JE, Manson JE, LaMotte F, Hennekens CH. Adherence to aspirin in the prevention of myocardial infarction: The Physicians’ Health Study. Arch Intern Med 1994; 154: 2649–2657.
18. Strom BL, ed. Pharmacoepidemiology. 2nd ed. New York: John Wiley and Sons, 1994.
19. Feinstein AR. Clinical Epidemiology: The Architecture of Clinical Research. Philadelphia: WB Saunders, 1985.
20. Walker AM. Confounding by indication. Epidemiology 1996; 7: 335–336.
21. Petrie H, Urquhart J. Channeling bias in the interpretation of drug effects. Stat Med 1991; 10: 577–581.
22. von Korff M, Wagner EH, Saunders K. A chronic disease score from automated pharmacy data. J Clin Epidemiol 1992; 45: 197–203.
23. Clark DO, von Korff M, Saunders K, Baluch WM, Simon GE. A chronic disease score with empirically derived weights. Med Care 1995; 33: 783–795.
24. Johnson RE, Hornbrook MC, Nichols GA. Replicating the Chronic Disease Score (CDS) from automated pharmacy data. J Clin Epidemiol 1994; 47: 1191–1199.
25. The Hearst Corporation. National Drug Data File (NDDF) User Manual. San Bruno, CA: The Hearst Corporation, 1988; 74–84.
26. Yuan Z, Cooper GS, Einstadter D, Cebul RD, Rimm AA. The association between hospital type and mortality and length of stay. Med Care 2000; 38: 231–245.
27. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chron Dis 1987; 40: 373–383.
28. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992; 45: 613–619.
29. Cox DR. Regression models and life tables (with discussion). J R Stat Soc Ser B 1972; 34: 187–220.
30. Mulrow CD, Cornell JA, Herrera CR, Kadri A, Farnett L, Aguilar C. Hypertension in the elderly: implications and generalizabiity of randomized trials. JAMA 1994; 272: 1932–1938.
31. Miettinen TA, Pyorala K, Olsson AG, Musliner TA, Cook TJ, Faergeman O, Berg K, Pedersen T, Kjekshus J. Cholesterol-lowering therapy in women and elderly patients with myocardial infarction or angina pectoris: findings from the Scandinavian Simvastatin Survival study (4S). Circulation 1997; 96: 4211–4215.
32. Lewis SJ, Moye LA, Sacks FM, Johnstone DE, Timmis G, Mitchell J, Limacher M, Kell S, Glasser SP, Grant J, Davis BR, Pfeffer MA, Braunwald E. Effect of pravastatin on cardiovascular events in older patients with myocardial infarction and cholesterol levels in the average range: results of the Cholesterol and Recurrent Events (CARE) trial. Ann Intern Med 1998; 129: 681–689.
33. Burt VL, Whelton P, Roccella EJ, Brown C, Cutler JA, Higgins M, Horan MJ, Labarthe D. Prevalence of hypertension in the US adult population: results from the Third National Health and Nutrition Examination Survey, 1988–1991. Hypertension 1995; 25: 305–313.
34. Solomon DH, Gurwitz JH. Toxicity of nonsteroidal anti-inflammatory drugs in the elderly: is advanced age a risk factor? Am J Med 1997; 102: 208–215.
35. Ray WA, Griffin MR, Downey W. Benzodiazepines of long and short elimination half-life and the risk of hip fracture. JAMA 1989; 262: 3303–3307.
36. Walker AM, Stampfer MJ. Observational studies of drug safety. Lancet 1996; 348: 489.
37. Psaty BM, Koepsell TD, Lin D, Weiss NS, Siscovick DS, Rosendaal FR, Pahor M, Furberg CD. Assessment and control for confounding by indication in observational studies. J Am Geriatr Soc 1999; 47: 749–754.
38. Marshall JR, Hastrup JL, Ross JS. Mismeasurement and the resonance of strong confounders: correlated errors. Am J Epidemiol 1999; 150: 88–96.

aging; bias; confounding; pharmacoepidemiology; prescribing

© 2001 Lippincott Williams & Wilkins, Inc.