Clinical trials themselves provide some opportunity to effectively screen at baseline and follow any number of comorbid conditions and gain a brief snapshot of the history of the disease. However, clinical trials by design seek relatively uncomplicated patients, randomize them to a medication, and capture information from the beginning of their medication "trial". If we really want to be able to compare the impact of an intervention based on clinical trials and observational data, we must identify an incident or inception cohort, i.e. a cohort of patients at a definitive point in the course of their illness that can be reliably identified across all patients or at the point of their initial intervention.
Observational data can produce a sense of the real spectrum of disease and provide a clinical context for the "clean" patients normally enrolled in a clinical trial program. This will aid in the anticipation of adverse event issues and may help direct product labeling. However, it does put additional demands on any observational data source but is consonant with the goal of true characterization of the users of the medication. It is not uncommon for the initial users of a new therapy to be those who have failed on a currently available therapy. This is especially true when a new therapy is initially promoted as "safer": clinicians are then motivated to place all their more problematic treatment failures, or those who have suffered side-effects, on the new therapy. These patients invariably have more complicated clinical histories and by capturing information from the true beginning of therapy, one is able to better ascertain earlier periods of risk that may be associated with the medication (including other risk factors that may be modified by the new medication) that may not be obvious in a cross-sectional approach.
Longitudinal data are critical to the planning and interpretation of clinical trials as it would be foolhardy to design a trial whose length is too short to detect the events of interest or is underpowered for lack of information about the frequency of an outcome. It would be difficult to design a clinical trial in asthma with the intent of reducing Emergency Department (ER) admissions if the trialists had little insight into how frequently ER visits occur or how to at least identify the characteristics of the patients who do use the ER most frequently for their asthma. Disease natural history is critical to determine the appropriateness and effectiveness of clinical interventions, especially for those illnesses whose measures of treatment effect are unclear. Research from the Rochester Epidemiology Project illustrates this point well in work on benign prostatic hypertrophy (Guess et al., 1995), again pointing to the relative value of the data and the data source.
For all of the advantages offered by longitudinal data, there are some issues to be aware of especially when using clinical data. The issues are well described elsewhere (Stang, 1998; Strom, 2000); however, they bear brief mention here. Paramount among the concerns of the scientist using longitudinal or any data for that matter, is to understand why and how these data were collected. Data that derive from clinical practice or billing will only represent detected clinical illness and may not contain other factors that are useful in assessing risk (e.g. non-prescription drug use (OTC, street drugs, alcohol, smoking) or other medical disorders that may not be clearly or consistently captured in the data). The lack of capture may be due to how data are captured for a segment of care (hospitalizations for example) which may lack the necessary detail (as in the aggregates of claims data) or rely on the practitioner to make a separate entry based on a referral letter (as in GPRD).
For many databases, it is also important to understand the reason why people are in the database and if that confers its own special risk. For instance, government entitlement programs (e.g. Medicaid) capture information on a population of relatively low socio-economic status which, as a comorbidity, confers its own risks. Eligibility is determined on a monthly basis so that one often finds an interrupted series of information about a given patient. Clinical databases, such as the GPRD and the Rochester Epidemiology Project, are based on the medical care provided to a relatively well-defined population, but again reflects their medical care and may capture some additional information about social and psychological risk factors. The reader is encouraged to explore the amassing literature on databases and their applications in this area.
An important feature that often eludes adequate study, in large part because of the paucity of longitudinal data, is the relevance and station of time in both the onset and progression of disease. Cancer and environmental epidemiologists have long studied the lag time between putative exposures and detection of disease. However, it is a difficult study when faced with charting the natural history of disease whose outcomes may take 15 or 20 years to present themselves. However, the importance of this information is critical as we strive to understand the risk, benefit and costs of treating the asymptomatic patient prior to the onset of any serious outcomes. In these instances we may rely on other study designs or case series to obtain information that may otherwise be difficult to find.
With longitudinal data come more complicated methodologies as time introduces a very complicated covariate. New techniques, such as data mining, are also being introduced more and more into the signal detection effort, which is expanding our knowledge in one respect while generating the need for more "context" in our effort to determine statistical from clinically meaningful patterns. Data mining is able to find relationships between factors that may have escaped our detection but these findings will still require a basis for evaluation.
Was this article helpful?
If you suffer with asthma, you will no doubt be familiar with the uncomfortable sensations as your bronchial tubes begin to narrow and your muscles around them start to tighten. A sticky mucus known as phlegm begins to produce and increase within your bronchial tubes and you begin to wheeze, cough and struggle to breathe.