Use of the case-crossover design to study prolonged drug exposures and insidious outcomes☆
Introduction
The continuing increase in use of therapeutic drugs makes it imperative to develop accurate and efficient methods to study potential adverse effects, especially in the elderly (1). Standard pharmaco-epidemiologic designs are often criticized for their vulnerability to confounding, arising from the absence or questionable accuracy of clinical information in claims data 2., 3.. For this reason, the development of the case-crossover study has been an important advance (4). In this method, exposure histories are obtained only on subjects who had outcomes (cases). Relative risks are then estimated by comparing the frequency of exposure immediately before the case-defining event (i.e., the case period) vs. an earlier period (i.e., the control period). Because cases are self-matched, control selection bias, study inefficiency, and confounding by time-invariant characteristics are all reduced 5., 6..
The case-crossover design was originally intended to study brief exposures that have immediate and transient effects (e.g., physical exertion, anger, cocaine use, or cellular telephone calls), and acute outcomes that have abrupt and obvious onsets (e.g., myocardial infarction or motor vehicle accidents) 4., 7., 8., 9., 10., 11.. Some investigators have used case-crossover methods to study exposures with more prolonged effects (e.g., high vs. low intensity β-agonist use) as well as outcomes with delayed or insidious onsets (e.g., hemorrhagic fever) 12., 13.. This usage has mainly involved lengthening the case and control time windows to accommodate longer hazard periods from exposures, longer durations of exposures, or longer induction periods for outcomes (see Appendix). However, the effect of lengthening time windows on observed results has not been rigorously examined. The most worrisome concern is that added time in lengthened case and control periods could increase confounding by variables that fluctuate over time (14).
The first aim of the current study was to examine how lengthening the case and control time windows affects the detection of both true risks from drugs with established toxic properties, as well as spurious risks from drugs without such properties. The outcome in the current study, central nervous system (CNS) dysfunction, is a common druginduced illness with important consequences in the elderly 15., 16., 17.. Because it often develops and progresses gradually, it is under-recognized clinically and difficult to study with standard pharmacoepidemiologic methods 18., 19., 20., 21., 22.. To shed light on whether case-crossover designs can identify true risks, we chose three active drug regimens which can cause CNS dysfunction (oral corticosteroids, digoxin, and opiates) 23., 24., 25., 26., 27.. Their apparent effects may be especially prolonged or delayed in pharmacoepidemiologic studies of the elderly employing claims data because of the altered pharmacokinetics/pharmacodynamics that occur with aging, delayed and intermittent consumption of filled prescriptions, as well as time lags between the onset of CNS dysfunction and the receipt of services through which cases are diagnosed (28). We also chose two inactive drug exposures that are generally without adverse CNS properties (multivitamins and statin lipid-lowering drugs), to investigate whether case-crossover designs with prolonged time windows are prone to identifying spurious drug effects (23).
The second aim was to evaluate whether lengthening the time window increases vulnerability to two types of confounding by temporally-varying characteristics. Confounding by comorbidity is an important possibility because many general medical conditions, including conditions underlying the use of the active regimens in this study, can cause CNS impairment in the elderly as well as affect prescribing decisions 29., 30.. Another specific form, confounding by contraindication, could occur if prescribers avoid CNS toxic drugs in vulnerable patients or discontinue them in patients with premonitory symptoms but not yet fully developed outcomes. Confounding by increased “medicalization” around the time of CNS impairment, which would occur if cases increased their utilization of all types of health care, including drugs, is also possible around the time of a gradually progressive outcome.
A third and related aim was to see if such case-crossover studies could be performed using claims data. Claims data have several potential advantages including their large size, secondary nature, and accurate information on drug and service use. If the potential disadvantage of absent or questionably accurate clinical information 2., 3. can be overcome, claims data might offer a timely and relatively inexpensive source of information for studies of drug-induced CNS dysfunction.
Section snippets
New Jersey medicaid program
We extracted information on all individuals enrolled in the New Jersey Medicaid program from January 1, 1990 to June 30, 1995, including: demographic information and dates of enrollment; outpatient, nursing home, and hospital utilization data; and data for all filled prescriptions. The indigent status of Medicaid enrollees results in essentially no out-of-pocket (i.e., out-of-system) health care utilization.
New Jersey pharmaceutical assistance to the aged and disabled ( P A A D) program
Additional information on non-indigent patients aged ⩾65 years, was derived for the same
Results
Characteristics of subjects (N = 8220) during the 3-month control time period (i.e., 93–182 days prior to the index date) are presented in Table 1. Nearly 1/3 of subjects were 85 years of age and older. The large majority were women, of white race, and non-indigent (i.e., eligible for PAAD as opposed to Medicaid). The most common specific diagnoses recorded during the control period were cardiovascular (i.e., hypertension, ischemic heart disease, and congestive heart failure). The utilization of
Conclusions
These results suggest that the case-crossover design maybe a useful method for the study of drug exposures with prolonged effects, and outcomes with delayed or insidious onsets. This application involved increasing the length of the case and control time periods during which drug exposures are assessed. Earlier studies of transient exposures with immediate effects and outcomes with short induction periods have typically used time windows lasting seconds to days 5., 6., 7., 8., 9., 10., 11..
References (38)
- et al.
Delirium and other organic mental disorders in a general hospital
Gen Hosp Psychiatry
(1985) - et al.
Under-recognized delirium in ED geriatric patients
Am J Emerg Med
(1995) - et al.
Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases
J Clin Epidemiol
(1992) - et al.
A new method of classification of prognostic comorbidity for longitudinal studies: development and validation
Journal of Chronic Diseases
(1987) The American health care system: expenditures
N Engl J Med
(1999)- et al.
Medicaid Data as a Source for Postmarketing Surveillance Information, Final Report
(1984) - et al.
The accuracy of Medicare's hospital claims data: Progress has been made, but problems remain
Am J Public Health
(1992) The case-crossover design: a method for studying transient effects on the risk of acute events
Am J Epidemiol
(1991)- et al.
Control sampling strategies for case-crossover studies: An assessment of relative efficiency
Am J Epidemiol
(1995) Confounding and exposure trends in case-crossover and case-time-control designs
Epidemiology
(1996)
Triggering of acute myocardial infarction by physical exertion: Protection against triggering by regular exertion
N Engl J Med
Triggering of acute myocardial infarction by episodes of anger
Circulation
Triggering of acute myocardial infarction by sexual activity: Low absolute risk and prevention by regular physical exertion
JAMA
Triggering of myocardial infarction by cocaine
Circulation
Association between cellular-telephone calls and motor vehicle collisions
N Engl J Med
The case-time-control design
Epidemiology
A comparison of case-crossover and case-control designs in a study of risk factors for hemorrhagic fever with renal syndrome
Epidemiology
Case-crossover and case-time-control designs as alternatives in pharmacoepidemiologic research
Pharmacoepidemiol Drug Saf
Neuropsychiatric syndromes in the elderly: An overview
Cited by (0)
- ☆
This work was supported by NIMH grant K01-MH01651 (to Dr. Wang), AHRQ grant R01-HS10881 (to Dr. Schneeweiss), NIA grants R01-AG18833 (to Dr. Glynn) and R03-AG18395 (to Dr. Avorn), and a research grant from Pharmacia Corporation.