No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Assessment of the "Case-Chaos" Design as an Adjunct to the Case-Control Design
Abstract
In 2012, a novel case series method dubbed the “case-chaos” design was proposed as an alternative to case-control studies,
whereby controls are artificially created by permutating the exposure information of the cases. Our aim in the current work
was to further evaluate the case-chaos method. Using a theoretical example of 2 risk factors, we demonstrated that the case-chaos
design yields risk estimations for which the odds ratios obtained for every risk factor are in the same ascending order as
the risk factors' exposure prevalences in the case group. Applying the method to data from the European Study of Severe Cutaneous
Adverse Reactions (EuroSCAR; 1997–2001), we were not able to obtain sensible results but instead produced results as predicted
by our theoretical assessment. We therefore claim that the method is equivalent to declaring risk solely on the basis of prevalences
obtained in cases. While the proposers of the case-chaos method view it as a useful adjunct, we show that it cannot produce
sensible estimates.
... As Edelstein et al. noted, case-chaos methodology is an approach to comparing exposure frequencies quantitatively and with a measure of uncertainty around these estimates. Doerken et al. (3) described the mathematical basis for the monotonic relationship between the case-chaos matched odds ratio and the exposure prevalence in a 2-exposure model. In that large case-chaos analysis with 379 cases in which they simulated 10 controls per case, falsely reassuring statistical precision was evident, which is in contrast to the outbreak and field epidemiology contexts that we proposed (2). ...
... We agree with Edelstein et al. that synthesis of background measures of exposure and specific incident investigation data are also useful adjuncts to classical analytical epidemiologic approaches and may outperform case-chaos analysis when such data are available. Doerken et al. also tested case-chaos methodology empirically on the association between the severe cutaneous adverse reactions and exposure to medicines (3). Their predictably negative result highlights the need to consider context. ...
... We thank the authors of these 2 studies for their contribution to mapping out the utility and limits of the case-chaos approach in practice, although both draw negative conclusions on the basis of 1 (3) and 5 (1) studies. Of the 10 studies to date (1)(2)(3)(4), the case-chaos approach identified the appropriate exposure in 7 studies and closely related exposures in 2, which we think represents a good return in practical public health terms. ...
... The EuroSCAR data was previously investigated for a different case series design called case-chaos (Doerken et al., 2014). It showed that the sparseness of the risk factors played a key role for the success and failure of the case-chaos design. ...
Modeling binary outcomes plays an important role in epidemiology. It is often used to study the influence of risk factors on the occurrence or onset of a disease. Logistic regression via maximum likelihood (ML) has become the established analysis method of choice. However, risk factors with very low prevalences are problematic. Risk estimation of such sparse covariates via ML often suffers from strong bias. Even though this challenge is common in epidemiology, investigations into appropriate methodology exists are few in number.
The dissertation is a treatise of solutions in this context via penalized regression. Penalizing the likelihood function makes estimation even of very sparse covariates possible when classical ML methods often fail. In recent years, penalized regression has received increased attention in statistical literature, and as a result, a number of different variations exist. However, these methods have received little to no attention in epidemiologic studies.
In this dissertation penalized methods will for the first time be investigated in the context of sparse covariates. Firstly, the theory concerning state-of-the-art penalization methods will be treated. Further, the investigated methods will be studied using data from epidemiological studies. Lastly, accompanying simulation studies are conceived and analyzed. Tuned extensions of existing penalization methods are proposed to improve their untuned counterparts.
... case-chaos methodology [10][11][12][13][14], two on case-case methodology [15,16], two on case-control methodology [17,18], one discussing the validity of case-cohort methodology [19] and one discussing the validity of case-crossover methodology [20]. The use of laboratory methods, including whole genome sequencing, was described in five (15.2%) papers [21][22][23][24][25]. Traceback procedures were explored in five (15.2%) papers, including three on the use of network analysis [26][27][28], one on the use of food flow information [29] and one examining the use of relational systems to identify sources common to different cases [30]. ...
Enteric illness outbreaks are complex events, therefore, outbreak investigators use many different hypothesis generation methods depending on the situation. This scoping review was conducted to describe methods used to generate a hypothesis during enteric illness outbreak investigations. The search included five databases and grey literature for articles published between 1 January 2000 and 2 May 2015. Relevance screening and article characterisation were conducted by two independent reviewers using pretested forms. There were 903 outbreaks that described hypothesis generation methods and 33 papers which focused on the evaluation of hypothesis generation methods. Common hypothesis generation methods described are analytic studies (64.8%), descriptive epidemiology (33.7%), food or environmental sampling (32.8%) and facility inspections (27.9%). The least common methods included the use of a single interviewer (0.4%) and investigation of outliers (0.4%). Most studies reported using two or more methods to generate hypotheses (81.2%), with 29.2% of studies reporting using four or more. The use of multiple different hypothesis generation methods both within and between outbreaks highlights the complexity of enteric illness outbreak investigations. Future research should examine the effectiveness of each method and the contexts for which each is most effective in efficiently leading to source identification.
... Furthermore, the limitations highlighted in the present study are inherent to the casechaos method rather than specific to applying it to outbreak investigations. These findings are therefore likely to be similar regardless of the type of data used, as highlighted in an evaluation of the case-chaos method using data from an adverse drug reaction case-control study (13). ...
Case-chaos methodology is a proposed alternative to case-control studies that simulates controls by randomly reshuffling the exposures of cases. We evaluated the method using data on outbreaks in Sweden. We identified 5 case-control studies from foodborne illness outbreaks that occurred between 2005 and 2012. Using case-chaos methodology, we calculated odds ratios 1,000 times for each exposure. We used the median as the point estimate and the 2.5th and 97.5th percentiles as the confidence interval. We compared case-chaos matched odds ratios with their respective case-control odds ratios in terms of statistical significance. Using Spearman's correlation, we estimated the correlation between matched odds ratios and the proportion of cases exposed to each exposure and quantified the relationship between the 2 using a normal linear mixed model. Each case-control study identified an outbreak vehicle (odds ratios = 4.9-45). Case-chaos methodology identified the outbreak vehicle 3 out of 5 times. It identified significant associations in 22 of 113 exposures that were not associated with outcome and 5 of 18 exposures that were significantly associated with outcome. Log matched odds ratios correlated with their respective proportion of cases exposed (Spearman ρ = 0.91) and increased significantly with the proportion of cases exposed (b = 0.054). Case-chaos methodology missed the outbreak source 2 of 5 times and identified spurious associations between a number of exposures and outcome. Measures of association correlated with the proportion of cases exposed. We recommended against using case-chaos analysis during outbreak investigations.
Standard cohort and case-control designs are suited to the study of cumulative effects of chronic exposures, but they are prone to confounding by indication. Case-crossover and case-time-control studies are especially useful for studying intermittent exposures with transient effects, and are less susceptible to confounding by indication. Each design has its strengths and weaknesses. Despite the increasing availability of automated databases, cohort studies are usually time consuming and expensive, and therefore not preferred for time-critical decisions. In case-control studies, the selection of appropriate controls can be difficult and time consuming, and sometimes impractical when the exposure is rare. Case-crossover studies use the exposure history of each case as his or her own control to examine the effect of transient exposures on acute events. It further allows to study the time relationship of immediate effects to the exposure. This design eliminates between-person confounding by constant characteristics, including chronic indications. Because exposure data for the case and control periods are provided by the same person, the problems of differential recall may be reduced in many but not all case-crossover studies. Bias can result from temporal changes in prescribing or within-person confounding, including transient indication or changes in disease severity. The case-time-control design is an elaboration of the case-crossover design, which uses exposure history data from a traditional control group to estimate and adjust for the bias from temporal changes in prescribing. This paper will present a structured decision table of when to use which design in pharmacoepidemiologic research. In conclusion, case-crossover and case-time-control studies are the designs of choice when separating acute effects from chronic effects of transient exposures and if confounding by indication is an outstanding problem.
Stevens-Johnson syndrome (SJS) and toxic epidermal necrolysis (TEN) are rare but severe cutaneous adverse reactions (SCAR) related to a variety of medications. They have a significant public health impact because of high mortality and morbidity. A multinational case-control study conducted in Europe between 1997 and 2001 evaluated the risk of medications to induce SCAR. Cases were actively detected through a hospital network covering more than 100 million inhabitants. Three hospitalized patients per case matched on age, gender, and date of interview were enrolled as controls. After validation by an expert committee blinded to exposures, 379 SCAR cases and 1,505 controls were included. Among drugs recently introduced into the market, strong associations were documented for nevirapine (relative risk (RR)>22) and lamotrigine (RR>14), and weaker associations for sertraline (RR=11 [2.7-46]), pantoprazole (RR=18 [3.9-85]), and tramadol (RR=20 [4.4-93]). Strong associations were confirmed for anti-infective sulfonamides, allopurinol, carbamazapine, phenobarbital, phenytoin, and oxicam-NSAIDs , with some changes in relative numbers of exposed cases. Thus, many cases were still related to a few "old" drugs with a known high risk. Risk was restricted to the first few weeks of drug intake. The use of such drugs as first-line therapies should be considered carefully, especially when safer alternative treatments exist. A number of widely used drugs did not show any risk for SJS and TEN.
Case-control studies are important in infectious disease epidemiology for rapidly identifying and controlling risks, but challenges, including the need for speed, can place practical restrictions on control selection and recruitment. The biased comparisons that result can hamper or, worse, mislead investigators. Following a 2009 outbreak of Shiga-like toxin-producing Escherichia coli O157 infection associated with a petting farm in southeast England, it was hypothesized that case behavior alone could be used to identify risks. Case-patients' exposures were randomized on a case-by-case basis, and the resulting permuted data were compared with the actual events preceding illness by conditional logistic regression analysis. There was good agreement between the risks identified by using our new method and the risks elicited in the original outbreak case-control studies. This was also the case in analysis of 2 further historical outbreaks. These initial findings suggest that the technique, which we have called the "case-chaos" technique, appeared to be useful in this setting. Analysis of simulated data supports this view. Circumventing the need for traditional control data has the potential to reduce outbreak investigation lead times, leading to earlier interventions and reduced morbidity and mortality. However, further validation is necessary, coupled with an awareness of limitations of the method.
To assess case-only designs for surveillance with administrative databases.
We reviewed literature on two designs that are observational analogs to crossover experiments: the self-controlled case series (SCCS) and the case-crossover (CCO) design.
SCCS views the 'experiment' prospectively, comparing outcome risks in windows with different exposures. CCO retrospectively compares exposure frequencies in case and control windows. The main strength of case-only designs is they entail self-controlled analyses that eliminate confounding and selection bias by time-invariant characteristics not recorded in healthcare databases. They also protect privacy and are computationally efficient, as they require fewer subjects and variables. They are better than cohort designs for investigating transient effects of accurately recorded preventive agents, for example, vaccines. They are problematic if timing of self-administration is sporadic and dissociated from dispensing times, for example, analgesics. They tend to have less exposure misclassification bias and time-varying confounding if exposures are brief. Standard SCCS designs are bidirectional (using time both before and after the first exposure event), so they are more susceptible than CCOs to reverse-causality bias, including immortal-time bias. This is true also for sequence symmetry analysis, a simplified SCCS. Unidirectional CCOs use only time before the outcome, so they are less affected by reverse causality but susceptible to exposure-trend bias. Modifications of SCCS and CCO partially deal with these biases. The head-to-head comparison of multiple products helps to control residual biases.
The case-only analyses of intermittent users complement the cohort analyses of prolonged users because their different biases compensate for one another.
A case-control design involving only cases may be used when brief exposure causes a transient change in risk of a rare acute-onset disease. The design resembles a retrospective nonrandomized crossover study but differs in having only a sample of the base population-time. The average incidence rate ratio for a hypothesized effect period following the exposure is estimable using the Mantel-Haenszel estimator. The duration of the effect period is assumed to be that which maximizes the rate ratio estimate. Self-matching of cases eliminates the threat of control-selection bias and increases efficiency. Pilot data from a study of myocardial infarction onset illustrate the control of within-individual confounding due to temporal association of exposures.
A method is described for estimating the relative incidence of clinical events in defined time intervals after vaccination compared to a control period using only data on cases. The method is derived from a Poisson cohort model by conditioning on the occurrence of an event and on vaccination histories. Methods of analysis for event-dependent vaccination histories and survival data are discussed. Asymptotic arguments suggest that the method retains high efficiency relative to the full cohort analysis under conditions which commonly apply to studies of vaccine safety.
Many cardiovascular drugs have been implicated as causes of depression. With the exception of beta-blockers, few have been studied in formal epidemiologic designs. I present a new approach to such analyses that effectively controls for confounders that are stable over time. I analyzed the exposure histories of 11,244 incident antidepressant users, using the Odense University PharmacoEpidemiologic Database. All persons starting both beta-blockers and antidepressants during a predefined period were identified. If beta-blockers do not cause depression, this particular population should show equal numbers of persons starting either drug first. A depression-provoking effect of beta-blockers would generate an excess of persons starting beta-blockers first, that is a nonsymmetrical distribution of prescription orders. Confounders causing the two drugs to be co-prescribed would rarely be expected to affect the symmetry. The initial screening showed nonsymmetrical prescription orders for a wide range of cardiovascular drugs. After adjustment for an increasing incidence of antidepressant prescribing, I found a depression-provoking effect only for angiotensin-converting enzyme (ACE) inhibitors (rate ratio = 1.29; 95% confidence interval = 1.08-1.56) and calcium channel blockers (rate ratio = 1.31; 95% confidence interval = 1.14-1.51). This prescription sequence symmetry analysis may be useful as a screening tool.
We propose a residential case-specular method for the study of wire codes and childhood cancer. The method compares the wire codes of case residences with the wire codes of identical residences (specular residences) located in a virtual situation in which the position of the residence or the position of the power line is switched around the center of the street. It is designed to discriminate between the magnetic field hypothesis, which postulates that childhood cancer is affected by magnetic fields and that wire codes are a proxy for magnetic fields, vs the neighborhood hypothesis, which postulates that childhood cancer is affected by some characteristics of the neighborhood other than magnetic fields and wire codes are a proxy for those characteristics. The method is based on several assumptions that we tested with 400 randomly selected residences. Under certain conditions, the method also may allow effect estimation without requiring the selection of controls and the potential biases that result from control selection. The method is applicable to both past and future studies.
This article is the first of two parts on pharmacoepidemiology, a relatively new science that explores drug efficacy or toxicity using large observational study designs. The number of pharmacoepidemiologic studies published in medical journals has increased, as these studies have explored drug-related questions that at times cannot be answered in randomized clinical trials. Four pharmacoepidemiologic study designs will be discussed that explore the association between a specific pharmacologic agent and a disease of interest: cohort studies, case-control studies, case-crossover studies, and case-time-control studies.
The case-crossover design has been widely used to study the association between short-term air pollution exposure and the risk of an acute adverse health event. The design uses cases only; for each individual case, exposure just before the event is compared with exposure at other control (or "referent") times. Time-invariant confounders are controlled by making within-subject comparisons. Even more important in the air pollution setting is that time-varying confounders can also be controlled by design by matching referents to the index time. The referent selection strategy is important for reasons in addition to control of confounding. The case-crossover design makes the implicit assumption that there is no trend in exposure across the referent times. In addition, the statistical method that is used-conditional logistic regression-is unbiased only with certain referent strategies. We review here the case-crossover literature in the air pollution context, focusing on key issues regarding referent selection. We conclude with a set of recommendations for choosing a referent strategy with air pollution exposure data. Specifically, we advocate the time-stratified approach to referent selection because it ensures unbiased conditional logistic regression estimates, avoids bias resulting from time trend in the exposure series, and can be tailored to match on specific time-varying confounders.
The authors reply [letter]
- Nd Mccarthy
- Ia Gillespie
- O Brien
McCarthy ND, Gillespie IA, O'Brien SJ. The authors reply
[letter]. Am J Epidemiol. 2013;177(9):1022.
International Agency for Research on Cancer
- N E Breslow
- N E Day
Breslow NE, Day NE. Statistical Methods in Cancer Research.
Vol. 1. The Analysis of Case-Control Studies. (IARC Scientific
Publication no. 32). Lyon, France: International Agency for
Research on Cancer; 1980.