• - Google Chrome

Intended for healthcare professionals

  • Access provided by Google Indexer
  • My email alerts
  • BMA member login
  • Username * Password * Forgot your log in details? Need to activate BMA Member Log In Log in via OpenAthens Log in via your institution

Home

Search form

  • Advanced search
  • Search responses
  • Search blogs
  • Nested case-control...

Nested case-control studies: advantages and disadvantages

  • Related content
  • Peer review
  • Philip Sedgwick , reader in medical statistics and medical education 1
  • 1 Centre for Medical and Healthcare Education, St George’s, University of London, London, UK
  • p.sedgwick{at}sgul.ac.uk

Researchers investigated whether antipsychotic drugs were associated with venous thromboembolism. A population based nested case-control study design was used. Data were taken from the UK QResearch primary care database consisting of 7 267 673 patients. Cases were adult patients with a first ever record of venous thromboembolism between 1 January 1996 and 1 July 2007. For each case, up to four controls were identified, matched by age, calendar time, sex, and practice. Exposure to antipsychotic drugs was assessed on the basis of prescriptions on, or during the 24 months before, the index date. 1

There were 25 532 eligible cases (15 975 with deep vein thrombosis and 9557 with pulmonary embolism) and 89 491 matched controls. The primary outcome was the odds ratios for venous thromboembolism associated with antipsychotic drugs adjusted for comorbidity and concomitant drug exposure. When adjusted using logistic regression to control for potential confounding, prescription of antipsychotic drugs in the previous 24 months was significantly associated with an increased occurrence of venous thromboembolism compared with non-use (odds ratio 1.32, 95% confidence interval 1.23 to 1.42). The researchers concluded that prescription of antipsychotic drugs was associated with venous thromboembolism in a large primary care population.

Which of the following statements, if any, are true?

a) The nested case-control study is a retrospective design

b) The study design minimised selection bias compared with a case-control study

c) Recall bias was minimised compared with a case-control study

d) Causality could be inferred from the association between prescription of antipsychotic drugs and venous thromboembolism

Statements a , b , and c are true, whereas d is false.

The aim of the study was to investigate whether prescription of antipsychotic drugs was associated with venous thromboembolism. A nested case-control study design was used. The study design was an observational one that incorporated the concept of the traditional case-control study within an established cohort. This design overcomes some of the disadvantages associated with case-control studies, 2 while incorporating some of the advantages of cohort studies. 3 4

Data for the study above were extracted from the UK QResearch primary care database, a computerised register of anonymised longitudinal medical records for patients registered at more than 500 UK general practices. Patient data were recorded prospectively, the database having been updated regularly as patients visited their GP. Cases were all adult patients in the register with a first ever record of venous thromboembolism between 1 January 1996 and 1 July 2007. There were 25 532 cases in total. For each case, up to four controls were identified from the register, matched by age, calendar time, sex, and practice. In total, 89 491 matched controls were obtained. Data relating to prescriptions for antipsychotic drugs on, or during the 24 months before, the index date were extracted for the cases and controls. The index date was the date in the register when venous thromboembolism was recorded for the case. The cases and controls were compared to ascertain whether exposure to prescription of antipsychotic drugs was more common in one group than in the other. Despite the data for the cases and controls being collected prospectively, the nested case-control study is described as retrospective ( a is true) because it involved looking back at events that had already taken place and been recorded in the register.

Selection bias is of particular concern in the traditional case-control study. Described in a previous question, 5 selection bias is the systematic difference between the study participants and the population they are meant to represent with respect to their characteristics, including demographics and morbidity. Cases and controls are often selected through convenience sampling. Cases are typically recruited from hospitals or general practices because they are convenient and easily accessible to researchers. Controls are often recruited from the same hospital clinics or general practices as the cases. Therefore, the selected cases may not be representative of the population of all cases. Equally, the controls might not be representative of otherwise healthy members of the population. The above nested case-control study was population based, with the QResearch primary care database incorporating a large proportion of the UK population. The cases and controls were selected from the database and therefore should be more representative of the population than those in a traditional case-control study. Hence, selection bias was minimised by using the nested case-control study design ( b is true).

The traditional case-control study involves participants recalling information about past exposure to risk factors after identification as a case or control. The study design is prone to recall bias, as described in a previous question. 6 Recall bias is the systematic difference between cases and controls in the accuracy of information recalled. Recall bias will exist if participants have selective preconceptions about the association between the disease and past exposure to the risk factor(s). Cases may, for example, recall information more accurately than controls, possibly because of an association with the disease or outcome. Although in the study above the cases and controls were identified retrospectively, the data for the QResearch primary care database were collected prospectively. Therefore, there was no reason for any systematic differences between groups of study participants in the accuracy of the information collected. Therefore, recall bias was minimised compared with a traditional case-control study ( c is true).

Not all of the patient records in the UK QResearch primary care database were used to explore the association between prescription of antipsychotic drugs and development of venous thromboembolism. A nested case-control study was used instead, with cases and controls matched on age, calendar time, sex, and practice. This was because it was statistically more efficient to control for the effects of age, calendar time, sex, and practice by matching cases and controls on these variables at the design stage, rather than controlling for their potential confounding effects when the data were analysed. The matching variables were considered to be important factors that could potentially confound the association between prescription of antipsychotic drugs and venous thromboembolism, but they were not of interest as potential risk factors in themselves. Matching in case-control studies has been described in a previous question. 7

Unlike a traditional case-control study, the data in the example above were recorded prospectively. Therefore, it was possible to determine whether prescription of antipsychotic drugs preceded the occurrence of venous thromboembolism. Nonetheless, only association, and not causation, can be inferred from the results of the above nested case-control study ( d is false)—that is, those people who were exposed to prescribed antipsychotic drugs were more likely to have developed venous thromboembolism. This is because the observed association between prescribed antipsychotic drugs and occurrence of venous thromboembolism may have been due to confounding. In particular, it was not possible to measure and then control for, through statistical analysis, all factors that may have affected the occurrence of venous thromboembolism.

The example above is typical of a nested case-control study; the health records for a group of patients that have already been collected and stored in an electronic database are used to explore the association between one or more risk factors and a disease or condition. The management of such databases means it is possible for a variety of studies to be undertaken, each investigating the risk factors associated with different diseases or outcomes. Nested case-control studies are therefore relatively inexpensive to perform. However, the major disadvantage of nested case-control studies is that not all pertinent risk factors are likely to have been recorded. Furthermore, because many different healthcare professionals will be involved in patient care, risk factors and outcome(s) will probably not have been measured with the same accuracy and consistency throughout. It may also be problematic if the diagnosis of the disease or outcome changes with time.

Cite this as: BMJ 2014;348:g1532

Competing interests: None declared.

  • ↵ Parker C, Coupland C, Hippisley-Cox J. Antipsychotic drugs and risk of venous thromboembolism: nested case-control study. BMJ 2010 ; 341 : c4245 . OpenUrl Abstract / FREE Full Text
  • ↵ Sedgwick P. Case-control studies: advantages and disadvantages. BMJ 2014 ; 348 : f7707 . OpenUrl CrossRef
  • ↵ Sedgwick P. Prospective cohort studies: advantages and disadvantages. BMJ 2013 ; 347 : f6726 . OpenUrl FREE Full Text
  • ↵ Sedgwick P. Retrospective cohort studies: advantages and disadvantages. BMJ 2014 ; 348 : g1072 . OpenUrl FREE Full Text
  • ↵ Sedgwick P. Selection bias versus allocation bias. BMJ 2013 ; 346 : f3345 . OpenUrl FREE Full Text
  • ↵ Sedgwick P. What is recall bias? BMJ 2012 ; 344 : e3519 . OpenUrl FREE Full Text
  • ↵ Sedgwick P. Why match in case-control studies? BMJ 2012 ; 344 : e691 . OpenUrl FREE Full Text

nested case control study advantages

  • Research article
  • Open access
  • Published: 06 June 2017

Methodologic considerations in the design and analysis of nested case-control studies: association between cytokines and postoperative delirium

  • Long H. Ngo 1 , 2 ,
  • Sharon K. Inouye 2 , 3 , 4 ,
  • Richard N. Jones 3 , 5 ,
  • Thomas G. Travison 2 , 3 , 4 ,
  • Towia A. Libermann 2 , 7 ,
  • Simon T. Dillon 7 ,
  • George A. Kuchel 8 ,
  • Sarinnapha M. Vasunilashorn 1 , 2 , 3 ,
  • David C. Alsop 2 , 6 &
  • Edward R. Marcantonio 1 , 2 , 3 , 4  

BMC Medical Research Methodology volume  17 , Article number:  88 ( 2017 ) Cite this article

11k Accesses

12 Citations

1 Altmetric

Metrics details

The nested case-control study (NCC) design within a prospective cohort study is used when outcome data are available for all subjects, but the exposure of interest has not been collected, and is difficult or prohibitively expensive to obtain for all subjects. A NCC analysis with good matching procedures yields estimates that are as efficient and unbiased as estimates from the full cohort study. We present methodological considerations in a matched NCC design and analysis, which include the choice of match algorithms, analysis methods to evaluate the association of exposures of interest with outcomes, and consideration of overmatching.

Matched, NCC design within a longitudinal observational prospective cohort study in the setting of two academic hospitals. Study participants are patients aged over 70 years who underwent scheduled major non-cardiac surgery. The primary outcome was postoperative delirium from in-hospital interviews and medical record review. The main exposure was IL-6 concentration (pg/ml) from blood sampled at three time points before delirium occurred. We used nonparametric signed ranked test to test for the median of the paired differences. We used conditional logistic regression to model the risk of IL-6 on delirium incidence. Simulation was used to generate a sample of cohort data on which unconditional multivariable logistic regression was used, and the results were compared to those of the conditional logistic regression. Partial R-square was used to assess the level of overmatching.

We found that the optimal match algorithm yielded more matched pairs than the greedy algorithm. The choice of analytic strategy—whether to consider measured cytokine levels as the predictor or outcome-- yielded inferences that have different clinical interpretations but similar levels of statistical significance. Estimation results from NCC design using conditional logistic regression, and from simulated cohort design using unconditional logistic regression, were similar. We found minimal evidence for overmatching.

Conclusions

Using a matched NCC approach introduces methodological challenges into the study design and data analysis. Nonetheless, with careful selection of the match algorithm, match factors, and analysis methods, this design is cost effective and, for our study, yields estimates that are similar to those from a prospective cohort study design.

Peer Review reports

Nested case-control study (NCC) design within a prospective cohort study is used when the outcome data are available for all subjects, but the exposure of interest has not been collected, and is difficult or prohibitively expensive to obtain for all subjects [ 1 , 2 , 3 ]. NCC is cost effective and can be done with or without matching in the selection of a subset of the controls. NCC analysis with good matching procedure yields estimates that are as efficient and unbiased as estimates from the full cohort study [ 2 ]. The origin of the NCC design came from the desire to reduce computational costs of collecting and analyzing data for all subjects in a cohort study. Mantel proposed to sample the controls randomly from a finite cohort, and originally called this design “synthetic” case-control study [ 4 ]. Subsequently the use of matching to select the controls allowed for the implementation of the conditional likelihood functions and the demonstration of asymptotic consistency and efficiency property of the risk ratio estimates [ 5 ]. NCC has been used in many biomarker studies where it is expensive to collect and process biological samples for all subjects in the cohort study. Recent applications of NCC include studies showing the effects of serum lipids and lipoproteins on breast cancer risk [ 6 ], urine semaphorin-3A on renal damage in hypertensive patients [ 7 ], DNA methylation markers on type-2 diabetes [ 8 ], and plasma cytokines and the risk of HIV type one [ 9 ].

In addition to being cost effective, NCC with a smaller sample size tends to be less computationally demanding than the analysis of the full cohort study. If the match procedure is carried out properly, and the selected controls are representative of the controls in the cohort study, then NCC loses little efficiency compared to the full cohort analysis [ 3 ]. NCC could offer better validity than the full cohort study because the match procedure allows for adjustment for both measured, and for unmeasured confounders [ 10 ].

At the crux of the NCC design is the quality of the match procedure and the appropriate analysis that accounts for the match design. Algorithms used in match procedures such as the greedy algorithm, propensity score algorithm, and optimal algorithm are some of the most often used in NCC studies. Theoretically, the optimal algorithm has been demonstrated to outperform the greedy algorithm at the expense of computational costs [ 11 ]. However, in the context of our clinical study where the number of needed matched pairs was smaller than 50, it was not clear how large the difference would be in the performance of these algorithms. In terms of the interpretation of the analysis results, in the case of a binary outcome and a binary exposure, one can compute the odds ratio for the outcome, or the odds ratio for the exposure, and these two odds ratio estimators have been shown to be equivalent [ 12 ]. However, in the case of a binary outcome, and a continuous exposure (such as in biomarker discovery studies where the exposure is the level of putative marker concentration in plasma), it is not clear that the outcome odds ratio is equivalent to the mean or median of the paired differences (between the case and control in a selected pair). For a NCC with matching, Cornfield [ 12 ], Mantel and Haenszel [ 13 ], Breslow [ 2 ], Rubin [ 14 ], Rothman and Greenland [ 15 ] have pointed out that the match algorithm, the match factors, and their association with the outcome and the exposure play a critical role in validity and efficiency. In addition, caution is needed to avoid overmatching, since this could introduce bias and inefficiency into the estimators.

As a case study to evaluate these issues, we used a clinical study that focused on estimating the association between cytokines and postoperative delirium [ 16 , 17 , 18 , 19 ], a common and serious clinical syndrome that is associated with a sudden decline in attention and cognition. The study used a large cohort of older adults undergoing non-cardiac surgery enrolled in the SAGES: Successful Aging after Elective Surgery study [ 20 , 21 ]. Because we planned to employ high cost, labor intensive biomarker discovery technologies such as Luminex multiplex cytokine panels and proteomics using mass spectrometry, it was realistic to only measure the biomarkers in a subset of the patients. Therefore, we chose a NCC design. Cases with delirium were matched to a subset of controls without delirium. Controls were chosen based on the match of six demographic and baseline clinical variables thought to be potential confounders of the cytokine/delirium association.

To address the methodological issues described above, we set out to answer the following questions in this paper: 1) How much better, in terms of the number of selected matched pairs, and the quality of the match in a selected pair, is the optimal match algorithm compared to the greedy algorithm, and the propensity score algorithm? 2) Should we treat postoperative delirium incidence as the outcome and report the odds ratio estimates from the conditional logistic regression models, or should we treat cytokine (specifically, IL-6) concentration as the outcome and report the median paired difference between the delirium cases and controls? 3) Compared to the full cohort analysis, in our case with simulated IL-6 for the full cohort, how efficient and valid would the estimates from the NCC study be (both odds ratios from the conditional logistic regression model, and the median paired difference analysis)? 4) Is there evidence for overmatching, and how should that be quantified and interpreted?

Description of study data

The SAGES study has been described in detail previously [ 20 , 21 ]. Briefly, the study enrolled 566 adults age ≥70 who were scheduled for major non-cardiac surgery. Demographics and baseline clinical information such as comorbid conditions and cognitive function were collected preoperatively. During hospitalization, patient’s delirium status was assessed during daily interviews using the Confusion Assessment Method (CAM) and chart review. Functional and cognitive data and other outcomes were collected at baseline and at each of the follow-up time point. Blood samples for each patient were collected prior to surgery (PREOP), in the post anesthesia care unit (PACU), 2 days after surgery (POD2). One of the aims of the project was the estimation of the association between cytokines and postoperative delirium. After testing several kits in our pilot work, we chose the Luminex high-sensitivity kit from R&D Systems Inc. with 12 inflammatory cytokines to obtain estimated concentrations (pg/ml) at each of these 3 time points. To evaluate the methodology used in our NCC design for this paper, we will use only IL-6 as a representative cytokine.

Definition of delirium case and control

A case was defined as having [delirium on POD2], or [delirium on POD1 and subsyndromal (partial) delirium on either POD2 or POD3]. This definition was used to ensure that the blood sample on POD2 was reflective of the delirious state. A control was defined as not having delirium or subsyndromal delirium on any hospital day. We also required cases and controls to have blood samples with no or only mild hemolysis (a condition that may contaminate the plasma and cause inaccurate laboratory measurements). Our recent published work has more details on this issue [ 22 ]. At the time of this analysis, there were 272 subjects, 49 met the case definition, and 143 met the control definition. From these, 39 matched pairs were selected.

Match factors in the case-control design

Six factors were judged to be potentially confounding the association between cytokines and delirium. These factors were largely selected based on prior literature, clinical experience of the investigators and the Program Project Operations Committee that oversees the study. The six factors were: 1) age at surgery, 2) gender, 3) vascular comorbidity (having any one of the six conditions: myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, diabetes, and diabetes with end organ damage), 4) surgery type (orthopedic, vascular, gastrointestinal), 5) presence of the ApoE ɛ4 allele, 6) baseline GCP (general cognitive performance score, a summary measure derived from a detailed neurocognitive battery [ 23 ]. Among the match factors, only age and GCP were continuous variables. The other 4 match variables were categorical. After discussion with the study team, we decided to create our match algorithms such that we required an identical value between the case and the control on the four categorical variables, and a difference (caliper) of no more than five years for age, and no more than five points for GCP. Table  1 shows the distribution of the six match factors before and after the match. Of the 49 eligible cases and 143 eligible controls, 39 match pairs were created. The match cohort has identical prevalence for case and control on four categorical match variables (gender, surgery type, vascular comorbidity, APOE ɛ4), and similar mean and standard deviation for age and baseline GCP (Table  1 ).

Issue 1. Performance of match algorithms

We first considered three candidate match algorithms: propensity score [ 24 ], greedy [ 25 ], and optimal [ 11 ]. We eliminated the propensity score approach because the match was required to be exact for 4 out of 6 factors, and the propensity score method cannot guarantee this outcome. We then evaluated the two remaining algorithms: the greedy match, and the optimal match algorithm. The greedy match algorithm is widely used and implemented in many statistical procedures for matching and is computationally faster than the optimal match. However, in terms of the match quality, that is, the degree of similarity (measured by absolute, Euclidean, or Mahalanobis distance on the match factors between the case and the control), the optimal match algorithm has been shown to outperform the greedy algorithm [ 11 ]. Also, the optimal algorithm theoretically will yield a greater number of matches than the greedy algorithm. We will now briefly illustrate the application of both the greedy and optimal algorithms in our study.

For each case, the greedy algorithm first evaluates each of the controls and measures the total distance (we used the absolute value) from the six match factors between the case and the control. The requirement for a match is to have identical values on four categorical variables (required distance of zero), and no more than a 5-unit difference on each of two continuous variables. Thus for a match to be successful, the total distance must be less than ten units. The best match would have a total distance of zero. The larger the distance on the continuous variables, the worse the quality of the match.

The greedy algorithm evaluates all the controls that meet this requirement, and selects the control that has the minimum distance to a case to form a match pair. Both the case and control for this match pair are then eliminated from the pool of match eligible cases and controls. The optimal match algorithm is similar to the greedy algorithm; however, once the match pair is formed, its case and control are not eliminated from the pool, but rather can be uncoupled and matched again if the total distance up to that point with a new control or a new case is smaller. For a large dataset, it is this reconsideration to attain minimum total distance that makes the optimal match algorithm more computationally consuming than the greedy match.

Given the small sample size, we also performed the match algorithm based on at 1:2 design (one case for two controls). We aimed to assess if the results in the 1:1 NCC would hold when the sample size gets larger.

Issue 2. Choice of exposure vs. outcome

In analyzing data from this study, we were faced with a choice of treating the IL-6 levels as the predictor and delirium as the outcome versus making delirium the predictor and IL-6 the outcome. The former would involve reporting the odds ratio of delirium per unit increase (pg/ml) of IL-6 by using conditional logistic regression at the 3 time points [ 26 , 27 ], while the latter would result in reporting the mean or median of the paired matched case-control differences in IL-6 levels. In our case, IL-6 distributions were not normally distributed and the nonparametric approach [ 28 ] was more appropriate; therefore, the median paired difference (MPD) would be used to test for the null hypothesis of MPD equal to zero using the signed rank test.

Earlier work by Cornfield [ 12 ], showed that in the case-control design, if the exposure is binary, then the exposure odds ratio is indeed equal to the disease odds ratio. This is also true for the NCC design. In other words, treating delirium as either the outcome or exposure in the analysis would yield identical odds ratio estimate. In our case with the exposure variable IL-6 being continuous, it is no longer true that the estimated MPD can be equated to the disease (delirium) odds ratio. The two estimates also carry different clinical interpretations. The MPD conveys a longitudinal change of IL-6 through time due to delirium; whereas, the odds ratios from conditional logistic models help to assess how IL-6 influences delirium risk across different time points. After considering both approaches, we elected to use the MPD in the paper [ 22 ]. We felt it would be more informative to show how the median levels of cytokines varied over time in both the delirium and non-delirium groups. Here in this paper, we present both analysis methods, and examine the differences in the two estimators and their interpretations.

Issue 3. Nested case-control versus cohort design and analysis

A second analytic issue was our interest in comparing the NCC results to those of the cohort study results assuming that the data for the cohort study design were available. Mantel and Haenszel [ 13 ], and Breslow [ 2 ] articulated that the data from case-control study design could be thought of as a random sample, based on the outcome data, from the prospective cohort design; therefore, one should expect the case-control design to yield similar estimates as the cohort design. The difference in the estimators between the case-control and cohort design could indicate bias, which could be attributed to sources such as bias selection of the sampled controls, and/or the match factors, and the match algorithm in the case-control design. We hypothesized that the NCC match analysis using conditional logistic regression would yield estimates which are similar to those of the simulated cohort analysis and the conclusion about the IL-6 effect would be the same in both analysis methods.

To address this issue, since we only measured IL-6 in the 39 pairs of cases and controls, we used simulation to generate the IL-6 data for the whole cohort of study participants who were eligible for the match ( N  = 192). In addition to the 114 subjects who did not get matched, we decided to simulate IL-6 data also for the 78 subjects in the NCC study so that all subjects would have the same probability of being assigned randomly an IL-6 measurement. Based on the time point, and delirium status of the subject, a randomly selected measured IL-6 from the 78 matched subjects with the corresponding time point and delirium status was assigned, with replacement. We ran the simulation once and created one simulated dataset because the pool of measured IL-6 values ( N  = 78) to sample from was smaller than the number of required simulated values ( N  = 192). Using sampling with replacement, repeated samples would have been quite similar, and therefore one simulated sample was felt to be sufficient. The cohort analysis using simulated IL-6 data used a multivariable unconditional logistic regression with the independent variables being IL-6 and the six match factors, and the dependent variable being the binary delirium outcome variable.

Issue 4. Assessment of overmatching

A known phenomenon of the match algorithm in the NCC design is called overmatching, which can introduce bias and inefficiency into the estimation of the case-control study. Breslow [ 1 ] reported an in-depth simulation for a number of scenarios using a single binary match factor, one binary outcome, and one binary exposure. When there is no association between the match factor and exposure, or when there is no association between the match factor and the outcome, there is no need to use matching and stratified analysis (such as conditional logistic regression). One can just use a random sample of the controls, and use a non-stratified analysis. In fact, if matching is used, then the estimated odds ratio would be biased toward the null, and the variance of the odds ratio estimate would be inefficient (that is larger than that of an estimator derived from a random sample of the controls). Another situation is when there is an association between the match factor and the exposure, and an association between the match factor and the outcome. Matching would select controls with exposure value similar to that of the cases, leading to bias toward the null. The magnitude of this bias increases as the association between the match factor and the exposure increases. If matching is not used in this case, but a random sample of the controls is selected, then the bias is in fact worse, and goes away from the null. Thus, in this situation, no analysis solution would be available to fix the bias issue [ 1 , 2 ].

For our study, it was impossible to assess the association between the exposure IL-6 and the outcome delirium, with the six match factors at the design phase of the NCC study, because we did not have IL-6 data. We selected the six match factors using strictly clinical experts’ input and review of the literature. After the NCC study design had been implemented, and IL-6 was measured, we now know that there is an association between the exposure (IL-6) and the outcome (delirium incidence) on POD2. We also know that the match factors are jointly associated with the outcome. Thus the degree of bias in the estimate of the association between IL-6 and delirium depends on the association between IL-6 and the match factors. That is, if the joint distribution of the six match factors is associated with IL-6 concentration, then the estimation of IL-6 effect on postoperative delirium in the conditional logistic regression model could be underestimated (biased toward the null).

The reason for this phenomenon can be illustrated with an example. Assume that one of the match factors, vascular comorbidity, is associated with IL-6. So those subjects with the presence of vascular comorbidity have higher level of IL-6 than those who do not have this comorbidity. Take a matched pair of a case and a control subject who both had vascular comorbidity. The values of IL-6 for this pair would both be high because vascular comorbidity is associated with IL-6. Take another pair who did not have vascular comorbidity. This pair would have both IL-6 values in the lower range relative to the pair with vascular comorbidity. Thus, the difference of IL-6 between members of the pair, for each of these two pairs, would be small, and yield small IL-6 difference. If vascular comorbidity was not associated with IL-6, then within a pair, the value of IL-6 could be high for one member and low for the other, leading to a larger difference, and thus a stronger effect of IL-6. This is the effect of overmatching.

To assess overmatching, we used only the data of the controls. Overmatching was possibly due to not a single match factor, but rather a joint distribution of 6 factors that would be required to be associated with the biomarkers to potentially cause the underestimation of IL-6 effect. To further evaluate whether this was the case, we used a general linear model where the dependent variable was IL-6 concentration, and the independent variables were the six match factors. The strength of the association between the joint distribution of the match factors and IL-6 was measured by the R-squared estimate. From this linear model, using partial R-squared estimates, we also decomposed the model R-squared into individual components which reflect the association of each match factor on IL-6.

All data management and analyses for this paper were carried out using the SAS software. For the simulation, we used procedure SURVEYSELECT in SAS/STAT software [ 29 ] with the method of unrestricted random sampling (URS) which allows selection of subjects with equal probability and with replacement.

For issue 1, evaluating the performance of the match algorithms, we illustrated in Fig.  1 , with just 2 cases and 2 controls, a theoretical exercise demonstrating how both algorithms select the controls, and how the optimal algorithm yielded more match pairs with better quality than the greedy algorithm. To further illustrate the property of the greedy vs. optimal match algorithms using our data, in Table  2 , we conducted an exercise where we varied the caliper of age (the difference of age in years between the case and control in the match pair) and GCP (the difference in GCP units between the case and control in the match pair) from one to five units to compare the performance of the greedy and optimal match algorithms. The optimal algorithm yielded more matched pairs than the greedy algorithm at caliper four (34 pairs vs 32 pairs), and at caliper five (39 pairs vs 34 pairs). If the same number of pairs was chosen by both algorithms, the optimal algorithm yielded higher quality pairs (with smaller mean distance, for example 2.05 versus 2.06 for caliper of 3). Note that we used a fixed caliper of five for the actual study, and did not vary it as we did in this exercise.

Illustration of the difference between greedy and optimal match algorithm. A numerical example is given here to demonstrate the theoretical properties of the greedy and optimal match algorithm

We also performed a 1:2 design to see if the observation we saw in the 1:1 design holds. Table  3 below shows the result for both the optimal, and greedy match algorithm. Each case is set to match to two controls, but this was not always possible. So when there were not two controls available, one control was chosen. As a result, we have some cases with two controls and some cases with only one control. For example, for caliper one for age, and GCP, a total of eight pairs was obtained from seven cases. Six of seven cases matched to one controls, but there was one case that matched to two controls. Note that when the caliper of age and GCP was tight, for example, as one or two or even three, it was harder to get a match, and therefore, there were more 1:1 match pairs (one case to one control) than 1:2 match pairs. As the caliper got wider as in five and six, it was easier to satisfy the match criteria, and so there were more 1:2 matches than 1:1 (e.g. at caliper five, for the optimal match, there were 24 1:2 matched pairs and 12 1:1 match pairs; and at caliper six, there were 28 1:2 matches versus 11 1:1 matches). Notice that beginning at caliper four, the optimal algorithm yielded more match pairs than the greedy algorithm; therefore, even with 1:2 design, our conclusion on the superiority of the optimal algorithm holds true, as in the case of 1:1 match design (Table  2 ). Of note, when we expanded the caliper width to six units for age and GCP, as in Table  3 shows, the result still holds, the optimal algorithm yielded five more match pairs than the greedy algorithm. On the basis of its demonstrated superiority in both quality and quantity of matches, we chose the optimal algorithm for our matched, NCC study design.

For issue 2, evaluating the choice of exposure versus outcome, we analyzed the data using conditional logistic regression for the observed IL-6, and nonparametrically using the signed rank test [ 28 ] on the median of paired differences (Table  4 ). Inferentially, from both methods, the differences between cases and controls were not statistically significant at PREOP, PACU, but were significant at POD2 for both methods ( p  = 0.005). The odds ratio estimate for POD2 was 1.02 (95% CI: 1.01–1.03), and the MPD was 50.44 pg/ml. While the level of statistical significance was the same, the two estimates convey different interpretations: one expressed a 2% increase in the odds of delirium incidence per one pg/ml increase in IL-6, and the other yielded an estimate of the population median difference of IL-6 between delirium cases and controls on POD2. Ultimately, it is reassuring that the two analytic approaches yielded the same conclusions in terms of a statistically significant association between IL-6 and delirium at POD2, despite yielding different effect measures with different clinical interpretations.

We also carried out a sensitivity analysis assessing the influence of IL-6 values in the right hand tail of the distribution. We focused on POD2 because this was the time period that we found significant IL-6 effect on postoperative delirium (Table  4 , the matched analysis with conditional logistic regression). The question here is if the inference would change if we did not include potentially influential large IL-6 values (e.g. above 90 th , 95 th percentile of the IL-6 distribution) in the modeling. At the other two periods, PREOP and PACU, the findings were highly non-significant. The distribution of IL-6 on POD2 has a mean of 109.7, SD = 81.2, median = 93.7, minimum = 3.98, maximum = 410.5. As shown in Table  4 , including all IL-6 data yielded the log OR estimate of 0.0154 (SE = 0.0054) and p -value = 0.005. When we only included data below the 90 th percentile of the distribution (cut off IL-6 at 192), the log OR estimate = 0.0150 (SE = 0.0061) and p -value = 0.0135. When we included data below the 95 th percentile (cut off IL-6 at 316), the log OR estimate = 0.0148 (SE = 0.0056) and p -value = 0.0087. So these estimates changed slightly and the significant effect of IL-6 remains.

For issue 3, evaluating the two types of analyses: nested case-control, and cohort design, we compared, in Table  4 , the results obtained from the NCC design vs. the simulated cohort design. The NCC study and the unmatched cohort analysis yielded the same conclusion: no significant effect of IL-6 on PREOP, PACU, and significant effect on POD2. The point estimates of the odds ratios, and 95% confidence intervals were almost identical for POD2 between the two analysis methods (odds ratio = 1.02, 1.01–1.03). Out of the three time points, one has lower standard errors in the match analysis (PREOP). From this simulated analysis, we concluded that our match NCC design with N  = 78 yielded nearly identical point estimates and had similar statistical efficiency as a more traditional cohort design with N  = 192.

For issue 4, from Table  5 , we found that the largest R-squared estimate was 22.30% from the PREOP period, which is equivalent to a correlation coefficient of 0.47. In the PACU too, gender has the largest partial R-squared estimate of 8.38%, which is about 0.29 for correlation. No variables exceeded 9% for partial R-squared estimation. Due to the relatively low magnitude of these individual factor partial R-squared estimates, we believe that the bias due to overmatching is not a major issue in our design and analysis.

In this paper, we discuss methodological issues related to the nested, matched case-control design, which is being increasingly used in biomarker discovery studies. We discussed the potential advantages of such a design, as well as the resultant complexities in analysis and interpretation of the results.

By using the NCC study design rather than a more traditional cohort design, we performed biomarker assays on only a portion of the cohort, resulting in a substantial cost savings. The tradeoff in the case-control design was the potential loss of efficiency and presence of bias in our estimation. These two potential drawbacks could come from the match algorithm employed, the approach used for modeling of the data, and from overmatching. We evaluated two candidate match algorithms and found that the optimal algorithm was superior to the greedy algorithm in yielding more match pairs with higher match quality. The interesting lesson that we learned is that with a match design of 1:2 or 1:3, we could increase the match pairs and thus statistical power to the analysis; however, the analysis could become more complicated due to the lack of independence among the match pairs. This dependency would have made our MPD analysis which uses nonparametric signed rank test invalid since the signed rank test requires independence.

Our evaluation of the IL-6 association with postoperative delirium needs careful clinical interpretation. We hypothesized that if the association between PREOP IL-6 and delirium case was statistically significant at type-I error of 0.05 then we would consider PREOP IL-6 a risk marker. This definition also applied to PACU IL-6.

Another issue we encountered was whether to use an analytic strategy in which delirium was the outcome or the predictor. We compared the strategies of using non-parametric signed rank test (delirium is the predictor and median IL-6 levels are the outcome) vs conditional logistic regression (IL-6 levels are the predictor and delirium is the outcome). We found similar results in terms of statistical significance, although we felt the former approach yielded more clinically meaningful effect estimates. We used simulation to evaluate the impact of whether a cohort study would have yielded similar results to our NCC study. Using unmatched multivariable logistic regression modeling, we found the results of this simulated cohort study to be very similar to those of the chosen NCC study design. Finally, to assess for overmatching, we also checked for the representativeness of our match sample in comparison to the pre-matched sample, and found that in both the controls, and cases, the post-matched sample of 78 subjects was quite similar to the pre-match sample of 193 subjects. In addition, we conducted a post-hoc evaluation of the correlation of the measured IL-6 levels with the joint effect of our six match variables in the sample of post-matched controls and found low partial r-square estimates for the individual match factor in each time period. This finding allows us to conclude that the likelihood of bias toward the null due to overmatching was not a major issue in the analysis. We also checked for the similarity in distribution of the match factors between the 39 controls after the match, and the 143 controls before the match and found the two samples to be similar in five of the six variables. Given the small sample size that we have, the statistical tests to compare the distributions between pre-match and post-match controls, and cases may not have sufficient power to detect statistical significance. Clinically, we think that of the six factors, only APOE e4 may show a clinically relevant difference between the two samples.

We also examined the overmatching issue and found the partial r-square estimates of the six match factors to be small. These analyses indicate that selection bias in the controls is a minor issue.

Past studies in the literature indicated that NCC design has been used widely; however, most focused on the application of the design. For delirium research, in particular in the area of biomarker study, we are not aware of any published study which explores the methodological issues of the NCC design. Therefore, this detailed assessment of the methodological issues in our NCC study design provides insights that will inform the design of future studies, particularly in the field of biomarker discovery.

Abbreviations

Apolipoprotein E gene

Confusion assessment method

Deoxyribonucleic acid

General cognitive performance

  • Interleukin-6

Median paired difference

Nest case-control study

Post anesthesia care unit

Picogram per milliliter

Postoperative at 1 month

Postoperative day 2

Postoperative day 3

Preoperative

Successful aging after elective surgery

Breslow N. Design and analysis of case-control studies. Annual Rev Pub Health. 1982;3:29–54.

Article   CAS   Google Scholar  

Breslow NE. Statistics in epidemiology: the case-control study. J Am Stat Assoc. 1996;91:14–28.

Article   CAS   PubMed   Google Scholar  

Langholz B. Case–control study, nested. Encyclopedia of Biostatistics. 2005;1:646-655.

Mantel N. Synthetic retrospective studies and related topics. Biometrics. 1973;29:479–86.

Goldstein L, Langholz B. Asymptotic theory for nested case-control sampling in the Cox regression model. Ann Stat. 1992;20:1903–28.

Article   Google Scholar  

Martin LJ, Melnichouk O, Huszti E, Connolly PW, Greenberg CV, Minkin S, et al. Serum lipids, lipoproteins, and risk of breast cancer: a nested case-control study using multiple time points. J Natl Cancer Inst. 2015;107:5.

Google Scholar  

Viazzi F, Ramesh G, Jayakumar C, Leoncini G, Garneri D, Pontremoli R. Increased urine semaphorin-3A is associated with renal damage in hypertensive patients with chronic kidney disease: a nested case–control study. J Nephrology. 2015;28:315–20.

Chambers JC, Loh M, Lehne B, Drong A, Kriebel J, Motta V, et al. Epigenome-wide association of DNA methylation markers in peripheral blood from Indian Asians and Europeans with incident type 2 diabetes: a nested case-control study. Lancet Diabetes Endocrinol. 2015;2:526–34.

Kahle EM, Bolton M, Hughes JP, Donnell D, Celum C, Lingappa JR, et al. Plasma cytokine levels and risk of HIV type 1 (HIV-1) transmission and acquisition: a nested case-control study among HIV-1-serodiscordant couples. J Infect Dis. 2015;211:1451–60.

Article   PubMed   Google Scholar  

Wacholder S, SIlverman DT, McLaughlin JK. Mandel JSl. Selection of controls in case-control studies: III. Design options. Am J Epidemiol. 1992;135:1042–50.

Rosenbaum PR. Optimal matching for observational studies. J Am Stat Assoc. 1989;84:1024–32.

Cornfield J. A method of estimating comparative rates from clinical data; applications to cancer of the lung, breast, and cervix. J Natl Cancer Inst. 1951;11:1269–75.

CAS   PubMed   Google Scholar  

Mantel N, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer Inst. 1959;22:719–48.

Rubin DB. Matched sampling for causal effects. New York, NY: Cambridge University Press; 2006.

Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. Philadelphia, PA: Lippincott Williams & Wilkins; 2008.

Marcantonio ER, Rudolph JL, Culley D, Crosby G, Alsop D, Inouye SK. Serum biomarkers for delirium. J Gerontol A Biol Sci Med Sci. 2006;61:1281–6.

de Rooij SE, van Munster BC, Korevaar JC, Levi M. Cytokines and acute phase response in delirium. J Psychosom Res. 2007;62:521–5.

Rudolph JL, Ramlawi B, Kuchel GA, McElhaney JE, Xie D, Sellke FW, et al. Chemokines are associated with delirium after cardiac surgery. J Gerontol A Biol Sci Med Sci. 2008;63:184–9.

Article   PubMed   PubMed Central   Google Scholar  

Girard TD, Ware LB, Bernard GR, Pandharipande PP, Thompson JL, Shintani AK, et al. Associations of markers of inflammation and coagulation with delirium during critical illness. Intensive Care Med. 2012;28:1965–73.

Schmitt EM, Marcantonio ER, Alsop DC, Jones RN, Rogers Jr SO, Fong TG, et al. Novel risk markers and long-term outcomes of delirium: the successful aging after elective surgery (SAGES) study design and methods. J Am Med Dir Assoc. 2012;13:818.

Schmitt EM, Saczynski JS, Kosar CM, Jones RN, Alsop DC, Fong TG, et al. The successful aging after elective surgery study: cohort description and data quality procedures. J Am Geriatr Soc. 2015;63:2463–71.

Vasunilashorn SM, Ngo L, Inouye SK, Libermann TA, Jones RN, Alsop DC, et al. Cytokines and postoperative delirium in older patients undergoing major elective surgery. J Gerontol A Biol Sci Med Sci. 2015;70:1289–95.

Jones RN, Rudolph JL, Inouye SK, Yang FM, Fong TG, Milberg WP, et al. Development of a unidimensional composite measure of neuropsychological functioning in older cardiac surgery patients with good measurement precision. J Clin Exp Neuropsychol. 2010;32:1041–9.

Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70:41–55.

Avis D, Burgess D, Steele JM. Probabilistic analysis of a greedy heuristic for Euclidean matching. Probability Eng Info Sci. 1988;2:143–56.

Pregibon D. Data analytic methods for matched case-control studies. Biometrics. 1984;40:639–51.

Stokes ME, Davis CS, Koch GG. Categorical data analysis using SAS, Third Edition. Cary, NC:SAS Institute Inc.; 2012.

Wilcoxon F. Individual comparisons by ranking methods. Biom Bull. 1945;1:80–3.

SAS Institute. SAS/STAT 12.1 User’s Guide: Procedure SURVEYSELECT. SAS Institute Inc.; 2012;95:8020-8092.

Download references

Acknowledgements

We would like to acknowledge the important comments from the editor and the two reviewers. Our paper has been greatly improved due to their thoughtful review and suggestions.

This work is funded by a program project grant (P01AG031720) and leadership grant (K07AG041835) from the National Institute on Aging (NIA), under the direction of Dr. Inouye. Dr. Marcantonio is a recipient of a Mid-career Investigator Award in Patient-Oriented Research (K24AG035075), and R01AG051658 and R01AG030618 from the NIA. Dr. Vasunilashorn is funded by the Charles A. King Trust Postdoctoral Research Fellowship Program, Bank of America, N.A., Co-Trustee and T32AG023480 from the NIA. The funding agencies had no role in the preparation of this manuscript and the authors retained full autonomy in the preparation of this manuscript.

Availability of data and materials

The datasets analyzed for the current study are available from the corresponding author on reasonable request.

Authors’ contributions

LN: Conception and design, Statistical Analysis, Interpretation of the data, Drafting manuscript, Critical revision of the manuscript. SI: Acquisition of data, Conception and design, Interpretation of the data, Critical revision of the manuscript. RJ: Interpretation of the data, Critical revision of the manuscript. TT: Interpretation of the data, Critical revision of the manuscript. TL: Acquisition of data, Interpretation of the data, Critical revision of the manuscript. SD: Acquisition of data, Interpretation of the data, Critical revision of the manuscript. GK: Acquisition of data, Interpretation of the data, Critical revision of the manuscript. SV: Interpretation of the data, Critical revision of the manuscript. DA: Interpretation of the data, Critical revision of the manuscript. EM: Conception and design, Acquisition of data, Interpretation of data, Critical revision of the manuscript, Obtained funding. All authors have read and approved the final version of the manuscript.

Competing interests

The authors declare that they have no competing interests.

Consent for publication

Not Applicable. No specific data from any particular person is reported in the paper.

Ethics approval and consent to participate

Written informed consent was obtained from study participants with the approval of the institutional review boards of the two participating academic hospitals (Beth Israel Deaconess Medical Center, Brigham and Women’s Hospital, Boston, MA), and the study coordinating center Hebrew SeniorLife (Boston, MA).

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Author information

Authors and affiliations.

Division of General Medicine and Primary Care, Beth Israel Deaconess Medical Center, 330 Brookline Ave, CO-203, MA 02215, Boston, Massachusetts, USA

Long H. Ngo, Sarinnapha M. Vasunilashorn & Edward R. Marcantonio

Harvard Medical School, Boston, Massachusetts, USA

Long H. Ngo, Sharon K. Inouye, Thomas G. Travison, Towia A. Libermann, Sarinnapha M. Vasunilashorn, David C. Alsop & Edward R. Marcantonio

Aging Brain Center, Institute for Aging Research, Hebrew Senior Life, Boston, Massachusetts, USA

Sharon K. Inouye, Richard N. Jones, Thomas G. Travison, Sarinnapha M. Vasunilashorn & Edward R. Marcantonio

Division of Gerontology, Beth Israel Deaconess Medical Center, Boston, Massachusetts, USA

Sharon K. Inouye, Thomas G. Travison & Edward R. Marcantonio

Department of Psychiatry and Human Behavior, Warren Alpert Medical School, Brown University, Providence, Rhode Island, USA

Richard N. Jones

Department of Radiology, Beth Israel Deaconess Medical Center, Boston, Massachusetts, USA

David C. Alsop

Beth Israel Deaconess Medical Center Genomics, Proteomics, Bioinformatics and Systems Biology Center, Division of Interdisciplinary Medicine and Biotechnology, Beth Israel Deaconess Medical Center, Boston, Massachusetts, USA

Towia A. Libermann & Simon T. Dillon

UConn Center on Aging, University of Connecticut Health Center, Farmington, Connecticut, USA

George A. Kuchel

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Long H. Ngo .

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License ( http://creativecommons.org/licenses/by/4.0/ ), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated.

Reprints and permissions

About this article

Cite this article.

Ngo, L.H., Inouye, S.K., Jones, R.N. et al. Methodologic considerations in the design and analysis of nested case-control studies: association between cytokines and postoperative delirium. BMC Med Res Methodol 17 , 88 (2017). https://doi.org/10.1186/s12874-017-0359-8

Download citation

Received : 24 August 2016

Accepted : 11 May 2017

Published : 06 June 2017

DOI : https://doi.org/10.1186/s12874-017-0359-8

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

  • Case-control
  • Greedy match
  • Optimal match
  • Conditional logistic regression

BMC Medical Research Methodology

ISSN: 1471-2288

nested case control study advantages

User Preferences

Content preview.

Arcu felis bibendum ut tristique et egestas quis:

  • Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris
  • Duis aute irure dolor in reprehenderit in voluptate
  • Excepteur sint occaecat cupidatat non proident

Keyboard Shortcuts

7.2 - advanced case-control designs, nested case-control study:.

This is a case-control study within a cohort study. At the beginning of the cohort study \((t_0)\), members of the cohort are assessed for risk factors. Cases and controls are identified subsequently at time \(t_1\). The control group is selected from the risk set (cohort members who do not meet the case definition at \(t_1\).) Typically, the nested case-control study is less than 20% of the parent cohort.

Advantages of nested case-control

  • Efficient – not all members of the parent cohort require diagnostic testing
  • Flexible – allows testing of hypotheses not anticipated when the cohort was drawn (at \(t_0\))
  • Reduces selection bias – cases and controls sampled from the same population
  • Reduces information bias – risk factor exposure can be assessed with investigator blind to case status

Disadvantages

  • Reduces power (from parent cohort) because of reduced sample size by 1/(c+1), where c = number of controls per case

Nested case-control studies can be matched , not matched , or counter-matched.

Matching cases to controls according to baseline measurements of one or several confounding variables is done to control for the effect from confounding variables. A counter-matched study, in contrast, is when we matched cases to controls who have a different baseline risk factor exposure level. The counter-matched study design is used to specifically assess the impact of this risk factor; it is especially good for assessing the potential interaction (effect modification!) of the secondary risk factor and the primary risk factor. Counter-matched controls are randomly selected from different strata of risk factor exposure levels in order to maximize variation in risk exposures among the controls. For example, in a study of the risk for bladder cancer from alcohol consumption, you might match cases to controls who smoke different amounts to see if the effect of smoking is only evident at a minimum level of exposure.

Example of a Nested Case-Control Study: Familial, psychiatric, and socioeconomic risk factors for suicide in young people: a nested case-control study . In a cohort study of risk factors for suicide, Agerbo et al. (2002), enrolled 496 young people who had committed suicide during 1981-97 in Denmark matched for sex, age, and time to 24,800 controls. Read how they matched each case to a representative random subsample of 50 people born the same year!

Advantages of the nested case-control design in diagnostic research

Affiliation.

  • 1 Julius Center for Health Sciences and Primary Care, University Medical Center, Utrecht, The Netherlands. [email protected]
  • PMID: 18644127
  • PMCID: PMC2500041
  • DOI: 10.1186/1471-2288-8-48

Background: Despite its benefits, it is uncommon to apply the nested case-control design in diagnostic research. We aim to show advantages of this design for diagnostic accuracy studies.

Methods: We used data from a full cross-sectional diagnostic study comprising a cohort of 1295 consecutive patients who were selected on their suspicion of having deep vein thrombosis (DVT). We draw nested case-control samples from the full study population with case:control ratios of 1:1, 1:2, 1:3 and 1:4 (per ratio 100 samples were taken). We calculated diagnostic accuracy estimates for two tests that are used to detect DVT in clinical practice.

Results: Estimates of diagnostic accuracy in the nested case-control samples were very similar to those in the full study population. For example, for each case:control ratio, the positive predictive value of the D-dimer test was 0.30 in the full study population and 0.30 in the nested case-control samples (median of the 100 samples). As expected, variability of the estimates decreased with increasing sample size.

Conclusion: Our findings support the view that the nested case-control study is a valid and efficient design for diagnostic studies and should also be (re)appraised in current guidelines on diagnostic accuracy research.

Publication types

  • Research Support, Non-U.S. Gov't
  • Case-Control Studies*
  • Cross-Sectional Studies
  • Research Design*
  • Sensitivity and Specificity
  • Venous Thrombosis / diagnosis*

nested case control study advantages

EP717 Module 5 - Epidemiologic Study Designs – Part 2:

Case-control studies.

  •   Page:
  •   1  
  • |   2  
  • |   3  
  • |   4  
  • |   5  
  • |   6  
  • |   7  

On This Page sidebar

A Nested Case-Control Study

Interpretation of the odds ratio, test yourself, recap of case-control design.

Learn More sidebar

Now consider a hypothetical prospective cohort study among 89,949 women in whom the investigators took blood samples and froze them at baseline for possible future use. After following the cohort for 12 years the investigators wanted to investigate a possible association between the pesticide DDT and breast cancer. Since they had frozen blood samples collected at baseline, they had the option of having the samples tested for DDT levels. If they had done this, the table below shows what they would have found.

If they had had this data, they could have calculated the risk ratio:

RR = (360/13,636) / (1,079/76,313) = 1.87

However, the cost of analyzing each sample for DDT was $20, and to analyze all of them would have cost close to $1.8 million. So, like the previous study, the exposure data was very costly.

Although this was a prospective cohort study, we could regard the cohort as a source population and conduct a case-control study drawing samples from the cohort . We could, for example, analyze the blood samples on all of the women who had developed breast cancer during the 12 year follow up and on 2,878 randomly selected samples from the women without breast cancer (i.e., twice as many controls as cases). This would be described as a nested case-control study , i.e., nested within a cohort study.

The results might have looked like this:

Odds Ratio = (a/c) / (b/d) = (360/1,079) / (432/2,446)

= 1.89 during the 12 year follow up study

So, they could achieve an odds ratio that is very close to what the risk ratio would have been at a much lower cost: (1,439+2,878) x $20 = $86,340.

The odds ratio is a legitimate measure of association, and, when the outcome of interest is uncommon, it provides a good estimate of what the risk ratio would have been if a cohort study had been possible. When looking at increasingly common outcomes, the odds ratio gives estimates that are more extreme than the risk ratio, i.e., further away from the null value. 

Not surprisingly, the interpretation of an odds is therefore similar to the interpretation of a risk ratio.

  • The null value (no difference) is 1.0.
  • Odds ratios > 1 suggest an increase in risk
  • Odds ratios < 1 suggest a decrease in risk

The odds ratio above would be interpreted as follows:

"Women with high DDT blood levels at baseline had 1.89 times the odds of developing breast cancer compared to women with low blood levels of DDT during the 12 year observation period."

Calculate the odds ratio for the association between playing video games and development of hypertension. Interpret the odds ratio you calculate in a sentence. See if you can do both of these correctly before looking at the answer.

return to top | previous page | next page

Content ©2021. All Rights Reserved. Date last modified: April 21, 2021. Wayne W. LaMorte, MD, PhD, MPH

IMAGES

  1. nested case control studies advantages and disadvantages

    nested case control study advantages

  2. PPT

    nested case control study advantages

  3. PPT

    nested case control study advantages

  4. 31+ Case Study Samples

    nested case control study advantages

  5. Advantages And Disadvantages Of Nested Case Control Study

    nested case control study advantages

  6. The flowchart of the nested case-control study

    nested case control study advantages

VIDEO

  1. 22 January 2024

  2. COVID19 Infection & Pulmonary Fibrosis

  3. Epidemiological studies made easy #neetpg #mnemonics

  4. case control study part 1 || PSM || @Sudarshan263

  5. Case control studies simplified

  6. USA Study Application Process

COMMENTS

  1. Nested case-control studies: advantages and disadvantages

    a) The nested case-control study is a retrospective design. b) The study design minimised selection bias compared with a case-control study. c) Recall bias was minimised compared with a case-control study. d) Causality could be inferred from the association between prescription of antipsychotic drugs and venous thromboembolism.

  2. A Practical Overview of Case-Control Studies in Clinical Practice

    The main advantages of a nested case-control study are as follows: (1) cost reduction and effort minimization, as only a fraction of the parent cohort requires the necessary outcome assessment; (2) reduced selection bias, as both case and control subjects are sampled from the same population; and (3) flexibility in analysis by allowing testing ...

  3. Advantages of the nested case-control design in diagnostic research

    Potential advantages of a nested case-control design in diagnostic research. The nested case-control study design can be advantageous over a full cross-sectional cohort design when actual disease prevalence in subjects suspected of a target condition is low, the index test is costly to perform, or if the index test is invasive and may lead to ...

  4. A Practical Overview of Case-Control Studies in Clinical Practice

    The main advantages of a nested case-control study are as follows: (1) cost reduction and effort minimization, as only a fraction of the parent cohort requires the necessary outcome assessment; (2) reduced selection bias, as both case and control subjects are sampled from the same population; and

  5. A New Comparison of Nested Case-Control and Case-Cohort Designs and

    For both nested case-control and case-cohort designs, inverse probability weighting methods were more powerful than the standard methods. However, the difference became negligible when the proportion of failure events was very low (<1%) in the full cohort. The comparison between two designs depended on the censoring types and incidence ...

  6. Advantages of the nested case-control design in diagnostic research

    Background Despite its benefits, it is uncommon to apply the nested case-control design in diagnostic research. We aim to show advantages of this design for diagnostic accuracy studies. Methods We used data from a full cross-sectional diagnostic study comprising a cohort of 1295 consecutive patients who were selected on their suspicion of having deep vein thrombosis (DVT). We draw nested case ...

  7. Analysis of Nested Case-Control Study Designs: Revisiting the Inverse

    An example of a nested case-control study design (This is an example of a nested case-control design (m= 2) from a small cohort of ten subjects. For example, three subjects (i = 1, 2, 3) failed with no ties before the remaining seven subjects (i = 4, …, 10) were censored. The subjects 2 and 4 were selected as the controls from the risk set at ...

  8. Methodologic considerations in the design and analysis of nested case

    The nested case-control study (NCC) design within a prospective cohort study is used when outcome data are available for all subjects, but the exposure of interest has not been collected, and is difficult or prohibitively expensive to obtain for all subjects. A NCC analysis with good matching procedures yields estimates that are as efficient and unbiased as estimates from the full cohort study.

  9. Nested case-control study

    A nested case-control (NCC) study is a variation of a case-control study in which cases and controls are drawn from the population in a fully enumerated cohort. [1] Usually, the exposure of interest is only measured among the cases and the selected controls. Thus the nested case-control study is more efficient than the full cohort design.

  10. A Practical Overview of Case-Control Studies in Clinical Practice

    Case-control studies are one of the major observational study designs for performing clinical research. The advantages of these study designs over other study designs are that they are relatively quick to perform, economical, and easy to design and implement. Case-control studies are particularly appropriate for studying disease outbreaks, rare ...

  11. A Nested Case-Control Study

    A Nested Case-Control Study. Suppose a prospective cohort study were conducted among almost 90,000 women for the purpose of studying the determinants of cancer and cardiovascular disease. After enrollment, the women provide baseline information on a host of exposures, and they also provide baseline blood and urine samples that are frozen for ...

  12. Nested case-control studies

    Abstract. The nested case-control study design (or the case-control in a cohort study) is described here and compared with other designs, including the classic case-control and cohort studies and the case-cohort study. In the nested case-control study, cases of a disease that occur in a defined cohort are identified and, for each, a specified ...

  13. Nested case-control studies: advantages and

    A nested case-control study design was used. The study design was an observational one that incorporated the concept of the traditional case-control study within an established cohort. This design overcomes some of the disadvantages associated with case-control studies, 2 while incorporating some of the advantages of cohort studies. 3 4

  14. 7.2

    Nested Case-Control Study: This is a case-control study within a cohort study. At the beginning of the cohort study ( t 0), members of the cohort are assessed for risk factors. Cases and controls are identified subsequently at time t 1. The control group is selected from the risk set (cohort members who do not meet the case definition at t 1 .)

  15. Design and analysis of nested case-control studies for recurrent events

    3.1 |. Nested case-control studies: univariate case. In this article, we develop methods for performing estimation and inference for the joint frailty model for recurrent events and a terminal event in contexts where complete data is not available - for example, it may be expensive, time-consuming, or otherwise infeasible to collect certain exposure or covariate measures on the full cohort.

  16. PDF Case-Cohort Studies vs Nested Case- Control Studies

    A nested case-control study design involves the selection of several healthy controls for each case, typically from those still under observation at the time when the case developed the disease [3]. However, nested case-control studies have some limitations: 1) Inefficiency due to the alignment of each selected control subject to its matched case.

  17. Nested case-control studies: advantages and disadvantages

    The present nested case-control study measured the relative risk of self-reported breast cancer associated with dietary phosphate intake over 10 annual visits in a cohort of middle-aged U.S ...

  18. A Practical Overview of Case-Control Studies in Clinical Practice

    The main advantages of a nested case-control study are as follows: (1) cost reduction and effort minimization, as only a fraction of the parent cohort requires the necessary outcome assessment; (2) reduced selection bias, as both case and control subjects are sampled from the same population; and (3) flexibility in analysis by allowing testing of a hypotheses in the future that is not ...

  19. Advantages of the nested case-control design in diagnostic research

    We aim to show advantages of this design for diagnostic accuracy studies. Methods: We used data from a full cross-sectional diagnostic study comprising a cohort of 1295 consecutive patients who were selected on their suspicion of having deep vein thrombosis (DVT). We draw nested case-control samples from the full study population with case ...

  20. Application of the matched nested case-control design to the secondary

    A nested case-control study is an efficient design that can be embedded within an existing cohort study or randomised trial. It has a number of advantages compared to the conventional case-control design, and has the potential to answer important research questions using untapped prospectively collected data.

  21. A Nested Case-Control Study

    A Nested Case-Control Study. Now consider a hypothetical prospective cohort study among 89,949 women in whom the investigators took blood samples and froze them at baseline for possible future use. After following the cohort for 12 years the investigators wanted to investigate a possible association between the pesticide DDT and breast cancer.

  22. Nested case-control studies: advantages and disadvantages

    It is concluded that prescription of antipsychotic drugs was associated with venous thromboembolism in a large primary care population in a population based nested case-control study design. Researchers investigated whether antipsychotic drugs were associated with venous thromboembolism. A population based nested case-control study design was used. Data were taken from the UK QResearch primary ...

  23. Animals

    The visceral white nodules disease in the internal organs of Larimichthys crocea has caused significant harm in the aquaculture of this species, with Pseudomonas plecoglossicida considered one of the core pathogens causing this disease. In this study, we designed three pairs of specific nested PCR primers targeting the sctU gene of P. plecoglossicida, a crucial component of the Type III ...

  24. Bias in full cohort and nested case-control studies?

    Fundamentally, a properly executed case-control study nested in a cohort is valid if the corresponding analysis of the full cohort is valid. The mathematics of the likelihoods are the same for both, 5 as Langholz and Richardson 1 point out, and the same software procedures work for both. The only salient difference between the two designs is ...