Case Control Study

Dr Abhijit P Pakhare

All India Institute of Medical Sciences, Bhopal

Need of Case-Control study

Cohort study

  • Risk

  • Non-event to event

  • Temporality

  • Incidence

  • Relative risk

  • Attributable risk

When longitudinal follow up is not possible

  • Long latency period of chronic diseases

    • Smoking ~ CVD, Lung cancer

    • Diet & exercise ~ Osteoporosis

  • Disease incidence ~ relatively few

    • but all to be followed

  • Very rare diseases- cohort way inefficient

Example-1: Atypical fractures

  • mid 2000s, ~ unusual form of femoral fracture in women

  • Bisphosphonates suspected because

    • newly introduced at that time

    • act by reducing bone remodeling.

  • Case series ~ bisphosphonates and atypical fractures,

    • other drugs

    • other diseases

  • Whether bisphosphonates were independently associated with atypical fractures?

Example-1: Atypical fractures

Example-1: Atypical fractures

Bisphonates use and atypical fractures of femoral shaft(Schilcher, Michaëlsson, and Aspenberg 2011)

  • National Swedish Patient Register,
  • all 59 women age 55 years or older with atypical femoral fractures in 2008
  • 263 controls, women in the same registry who had had ordinary femoral fractures
  • other variables - age, use of bone-modifying drugs (corticosteroids or estrogens), and diseases such as osteoporosis and previous fractures
  • concluded women taking bisphosphonates were 33 times more likely to develop atypical fractures

Design of Case-Control Study

Flow of Case-Control Study

Control

  • Control in other situations

  • Experimental studies

    • those who are not exposed to the intervention

  • Diagnostic laboratories

    • Specimen which have known amount of what is being tested

  • Case-control study-

    • those who do not have disease or outcome

    • controls- take in account or subtract the effects of other variables

Selection of cases

  • Diagnosis - rigorously confirmed

  • All cases or representative sample from defined population

  • Hospital cases

    • Severity or atypical nature

    • Catchment area

Selection of controls

  • Validity- comparability of cases and controls

  • Cases and controls should be members of same base population

  • Random sample of non-cases

  • Methods of selection

    • Population approach

    • Cohort approach

    • Hospital & Community approach

Population approach

  • Random sample of defined population

  • Dynamic population

  • Cases and controls selected at similar calender time

Cohort approach

  • Cases and control sampled from same cohort

  • All cases and random sample from base cohort

    • Case-cohort study

  • All cases and random sample at the time of incident case

    • Nested Case-Control study

Why do case-control within cohort?

  • When variables not available in cohort base are to be studied

  • not collected at base

    • not feasible to collect for all participants at baseline
    • cost

      • additional biomarkers

      • genetic analysis

    • time

      • missing information from records

      • additional questionnaire

Hospital and Community Controls

Hospital

  • Select comparable controls

  • Cases from hospital ward- controls from a different disease from same hospital

  • Hospitalized patients usually have different distribution compared to the rest of the population

Community

  • From catchment area of hospital

  • Neighborhood controls

Measurement of exposure

  • More than one control group

    • Community

    • Hospital

      • Another disease

      • Non-diseased - Caregivers

  • Odds ratios across groups

    • Similar- less likelihood of bias
    • Heterogeneous- Bias likely

Multiple controls per case

  • Different from multiple control groups

  • More than 1 control for a cases

  • Useful in case of very rare disease as cases will be few

  • To improve statistical power

  • Maximum up to 3 to 4 controls per case

Measurement of exposure

  • Complete & accurate records before development of disease

  • Can be done for prescriptions, medical records of investigations, procedures or done on stored samples

  • Behavioral exposures

  • Diet, exercise, substance use etc Disease status may influence recall

  • Exposure measurement methods and tools to be same for both cases and controls

Analysis of Case-Control Study

Comparisons

Cohort Study

  • We know totals of exposed and non-exposed since we only fix them

  • We estimate incidence in exposed & non-exposed and compare relative incidence

Case-Control Study

  • We know totals of Cases and Non-cases since we only fix them

  • We estimate exposure odds among cases & non-cases & compute ratio of two odds

Odds ratio

\(Odds\ of\ exposure\ in\ Cases = \frac{Number\ of\ exposed\ among\ Cases}{Number\ of\ non-exposed\ among\ Cases}\)

\(Odds\ of\ exposure\ in\ Cases = \frac{a}{c}\)

\(Odds\ of\ exposure\ in\ Non-Cases = \frac{Number\ of\ exposed\ among\ Non-Cases}{Number\ of\ non-exposed\ among\ Non-Cases}\)


\(Odds\ of\ exposure\ in\ Non-Cases = \frac{b}{c}\)

\(Odds\ ratio = \frac{Odds\ of\ exposure\ among\ Cases}{Odds\ of\ exposure\ among\ Non-Cases}\)

\(Odds\ ratio = \frac{ad}{bc}\)

Interpretation of Odds Ratio

  • Association of exposure and outcome

  • OR > 1 (more than one)

    • Cases are more likely to be exposed than non-cases. Exposure may be a risk factor

  • OR < 1 (less than one)

    • Cases are less likely to be exposed than non-cases. Exposure may be a protective factor

  • OR =1 (nearly one)

    • Exposure variable is similarly distributed among Cases and Non-Cases

Example: Bisphosnates and Atypical Fracture

             Outcome +    Outcome -      Total        Prevalence *        Odds
Exposed +           46           26         72                63.9      1.7692
Exposed -           13          237        250                 5.2      0.0549
Total               59          263        322                18.3      0.2243

Point estimates and 95% CIs:
-------------------------------------------------------------------
Odds ratio                                     32.25 (15.44, 67.39)
Attrib fraction (est) in the exposed (%)      96.84 (93.16, 98.62)
Attrib fraction (est) in the population (%)   75.55 (60.42, 84.90)
-------------------------------------------------------------------
Uncorrected chi2 test that OR = 1: chi2(1) = 128.657 Pr>chi2 = <0.001
Fisher exact test that OR = 1: Pr>chi2 = <0.001
 Wald confidence limits
 CI: confidence interval
 * Outcomes per 100 population units

Issues in Case-Control Study

Selection of cases

  • Prevalent cases

    • Associated factors for duration, survivors

  • Hospitals

    • Severity

    • Socio-economic position

  • Differential selection probability

Selection of controls

  • Hospital

    • Distinct distribution of variables than population

  • Household / Best friend

    • Over-matching

  • Population

    • Non-response

Measurement of exposure

  • Poor or limited recall

    • Past exposures not remembered well hence not reported
    • Non-differential misclassification

  • Differential recall

    • Diseased or cases easily identify with exposure than non-cases

    • Differential misclassification

Sources of presentation

Chapter-6 Risk: From disease to exposure (Fletcher, Fletcher, and Msc 2013)

Chapter- 7 & 9 : Observational studies (Celentano and Szklo 2018)

Thank You

References

Study Designs Workshop @ AIIMS Bhopal

Celentano, David D., and Moyses Szklo. 2018. Gordis Epidemiology. Elsevier Health Sciences.
Fletcher, Robert, Suzanne W. Fletcher, and Suzanne W. Fletcher, MD Msc. 2013. Clinical Epidemiology: The Essentials. Lippincott Williams & Wilkins.
Schilcher, Jörg, Karl Michaëlsson, and Per Aspenberg. 2011. “Bisphosphonate Use and Atypical Fractures of the Femoral Shaft.” New England Journal of Medicine 364 (18): 1728–37. https://doi.org/10.1056/NEJMoa1010650.