Skip to content

Embedding clinical interventions in observational cohort studies

November 6, 2012

Jigsaw PiecesAn observational cohort study, with well-characterized participants, comprehensive event identification, and extensive infrastructure, may offer a platform to conduct clinical trials or community interventions.  At the same time, an intervention could impact the data collection, analysis, and interpretation of longitudinal studies.

What are the advantages and disadvantages of this hybrid design? What types of clinical or community interventions would be best in this scenario?  How can epidemiology and clinical trial expertise be successfully integrated?  Are there budgetary benefits to embedding interventions? How would participants and communities react to this model?

We are seeking input on these questions as we consider new research directions.  We welcome your comments.

Posted by the Epidemiology Branch, NHLBI

4 Comments leave one →
  1. JohnDoe permalink
    November 9, 2012 4:33 pm

    Regarding the analysis, how we’d handle information about interventions is not particularly different to how we handle other covariates; either limit, adjust, or weight analyses to reflect the differences between the population you have in the (sub)study and the population about which you would like to make inferences. Epidemiologists should be used to these ideas.

    Epidemiologists might also note that, with well-documented and well-justified interventions, we know exactly why the intervention was made – including when it’s assigned randomly. Particularly with random assignment, the adjustments (or weighting, or whatever) are *easier* to make than for more familiar non-randomized covariates; fundamentally, this is why clinical trialists can (in some common situations) not worry as much about covariates as one might in an observational study.

    To facilitate analyses after an intervention has been embedded, studies would have to keep track of the intervention information, and make it available to everyone, together with recommendations/guidelines/instruction on how to do such analyses. This is a fairly large overhead, requiring considerable institutional memory and a deep understanding of the original study, but co-ordinating centers are good at providing these.

  2. Lewis H. Kuller, MD, DrPH permalink
    November 12, 2012 3:23 pm

    An intervention study should be considered a longitudinal study with better control of selected variables. Many coronary heart disease (CHD) longitudinal epidemiological studies are plagued by both the variability and high frequency of treatment of key risk factors, such as lipid lowering, treatment of BP, diabetes, use of aspirin, etc., variations in physical exercise, dieting to lose weight, weight loss drugs, etc. Use of these therapies in observational studies is not random and can have major effects on outcomes in relating risk factors to disease. They therefore can often both confound and bias the analysis. The second major problem is that it is extremely difficult to do good trials once observational studies have documented the strong association between a modifiable risk factor or technology and outcome. Thus, even if a technology has limited value or substantial value, the widespread use of the technology once observational studies have documented strong associations with an outcome, not necessarily causal, greatly limit subsequent clinical trials. The best example perhaps is the recent PLCO trial evaluating PSA, in which the use of PSA screening in the comparison group was so high as to probably invalidate the entire trial and perhaps misinterpret the results and subsequent policy implications.

    In the cardiovascular area, the hot debate has been coronary calcium measurement by CT, especially in primary prevention. As some of you know, I personally believe that such a trial is no longer feasible and probably unethical because of the extensive observational data providing the value of the screening technology and the availability of powerful lipid lowering drug therapies. Thus, after the initial coronary calcium studies, all of which were consistent with previous studies documenting that atherosclerosis is the primary determinant of clinical cardiovascular disease (CVD), plus the other studies of subclinical measurements, it would have been prudent to have done a clinical trial rather than strictly a observational MESA study. The results of MESA are consistent with all of the previous data regarding the value of measuring atherosclerosis. However, after the MESA results plus numerous other studies, the ability to randomize to a coronary calcium trial and subsequent therapies is practically impossible and most likely unethical.

    Observational studies have been embedded within the clinical trials, perhaps the best example was the use of the 360,000 screenees in the Multiple Risk Factor Intervention Trial for long term follow up which documented the powerful interrelationship between cholesterol, BP, smoking and history of diabetes and subsequent morbidity and mortality in Black, White and Hispanic men, especially younger age groups and with long term follow up. This was both a cheap and simple way of doing an evaluation of known risk factors utilizing the clinical trial screenees. Unfortunately, when these measures are done as part of a clinical trial, i.e. baseline, an effort to moderate costs often results in important information not being collected. For example, in the MRFIT no DNA was collected or stored cells and even height and weight was not measures because it was thought to be too costly. The Women’s Health Initiative has also evolved into several important and interesting observational studies, not only in CVD but also cancer, dementia, etc.

    The Ginkgo Evaluation of Memory Study (GEMS) was supposed to utilize the participants who were in the observational Cardiovascular Health Study. However, in spite of reported willingness by the participants prior to recruitment for the trial, very few actually pariticpated in the clinical trial. This is a warning of potential difficulties in getting participants in observationasl studies into a clinical trial, especially if they have been in the observational study for many years and also have not been seen in for clinic examinations over a long period of time. It would be interesting to see the response percentages in the WHI of potentially newly embedded clinical trials. It would probably be very useful to evlaute a variety of different methods to recruit individuals from an observational study like WHI into clinical trials and determine the actual yield for the trials and the success of keeping such individuals in the trials before expanding such efforts and wasting a fair amount of resources.

    An important potential value of using observational cohorts for clinical trials is the availability of the baseline and follow up data that can be used for selection for the clinical trial. For example, in the WHI it is possible to select women on a combination of poor risk factors, such as smoking, hypertension, diabetes and high cholesterol, and have CHD risk between 5/1,000 and 72/1000 person-years. Similarly, cognitive measures over time in the WHI can be used to select individuals in future trials related to Alzhemier’s disease, dementia, etc.

    It makes much more sense to consider both doing a clinical trial within an ongoing observational study as well as doing a trial rather than a large observational study and embedding in the trial enough information so that important new variables can be collected as well as stored specimens for future analysis. For example, in the recently reported prevalence of major cardiovascular diseases among Hispanics and Latinos, a large observational study is extraordinarily unlikely to provide any new information that was previously unknown. Why not randomize the participants into an aggressive pharmacological or nonpharmacological intervention versus perhaps a nonpharmacological intervention alone and a “usual follow up” and determine whether you can successfully reduce cardiovascular morbidity and mortality and, at the same time, determine the interrelationship of these numerous variables to both the outcomes, efficacies of the therapies and perhaps by controlling these other variables which are going to be difficult to control in the observational studies, be able to identify potentially new markers and link this with potential genetic analysis.

  3. February 16, 2013 6:43 pm

    The study populations in NIH NHLBI’s cohorts provide a unique resource. The participants have been studied comprehensively and a great deal is known about the cohorts. With the rapid influx of new technology for data collection, patient recruitment, and communication, the NIH NHLBI cohorts might serve as laboratories to explore new technologies and validate new methods. By studying methods within the NIH NHLBI cohorts, we will be better able to deploy these methods in other study populations. An example of such technology might be EpiCollect, a data collection approach using smartphones linked to web applications to collect data (Aanensen et al., 2009, PLOS One 4(9):e6968). One might pilot the new technology on a sub-sample of a NIH NHLBI cohort, to validate, understand and improve the technology for questions such as ease of data collection, reliability, accuracy, repeatability, and participant satisfaction. With the constant appearance of new technologies, the cohort populations might make a great contribution and help the research community focus in on the most valuable and effective technologies.

  4. March 13, 2013 4:48 pm

    Just saw this post..An international example that was part of a lecture here at JHSPH: The Rakai Health Sciences Program in the Rakai District, Uganda. The program has had an intermixing of both cohort and intervention components over the years, including serving as the setting for establishing male circumcision as a preventive procedure for HIV acquisition. See for more details.

Leave a Comment

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s