February 2017 Reading Notes

Journal Club

Spitzer, W. O.; Lewis, M. A.; Heinemann, L. A. J. & Thorogood, M. Third generation oral contraceptives and risk of venous thromboembolic disorders: an international case-control study BMJ, BMJ Publishing Group Ltd, 1996, 312, 83-88

Discussion Notes

Quick Hits

  • This 1996 multi-center matched case-control study found a slightly increased odds (OR 1.5) of exposure to third generation oral contraceptives (gestodene and desogestrel) compared to second generation oral contraceptives in VTE patients (471 cases, 1772 controls).
  • All case-control studies are prone to recall bias whereby cases are more likely to recall exposure than controls, which increases the likelihood of detecting an association. This is less likely to be an issue with an exposure as common as oral contraceptive pill use. *

Kevin Summary

Study Quality

Comments on validity (internal and external).
What would make for good controls?

Would this change clinical management?

Other key insights

  • OCPs prevent pregnancy - a higher VTE risk scenario
  • *

"Critical" Values

  • In 285 female VTE patients aged 18-50: odds of being pregnant was 4.6-fold that of controls, odds of being within 3 months postpartum was 60-fold that of controls. 1
  • Pregnancy incidence of 100 VTE events per 100,000 woman-years as compared to non-pregnant incidence of 23.3 VTE events per 100,000 woman-years. 4 to 5-fold increased risk (relative risk). 2
  • Death from pulmonary embolism in pregnancies occurs at a rate of 1.1 - 1.5 in 100,000 pregnancies 3

  • VTE risk on OCPs, 40-50 in 100,000 woman-years

Approach to OCPs

Outline

Authors and Funding source

A group of authors from McGill, Center for Epidemiology and Health Research in Germany, London School of Hygiene and Tropical Medicine, and Charing Cross and Westminster Medical School prepared the manuscript on behalf of the Transnational Research Group on Oral Contraceptives and the Health of Young Women who were requested to conduct the study by German regulatory authorities in response to concern about gestodene.

Funded by Schering AG Berlin (bought by Bayer in 2006, renamed Bayer Schering Pharma AG, renamed Bayer HealthCare Pharmaceuticals in 2011 per Wikipedia.org)

Research Question

Is there increased VTE risk with third generation oral contraceptives (i.e. ones containing gestodene or desogestrel)?

Study Design

Matched case-control.

Study Subjects

Cases: 471 women, aged 16-44

Controls: 1772 women matched on age and hospital versus community setting

16 centers in 5 countries

British and French community controls were by group general practice, and German/Austria/Switzerland ones were by neighborhood

Age matching was within 5 year bands

Predictor Variables

Exposure to any oral contraceptive, exposure to second generation oral contraceptive, exposure to third generation oral contraceptive

Defined as any use of oral contraceptives in 10 weeks prior to to diagnosis of CVD in controls.

Outcome Variables

Venous thromboembolism (i.e. PE and DVT)

Analysis

Stratified analysis and unconditional logistic regression with simultaneous adjustment.

Additional stratified analyses to search for unsuspected confounding variables and potential biases.

Conditional logistic regression to assess for overmatching.

Results

Adjusted odds ratios with 95% confidence intervals

All oral contraceptives vs current use 4.0 (3.1 - 5.3)
Third generation vs second generation 1.5 (1.1 - 2.1)
Germany 1.8
Britain 1.5

stratified by age and first use versus subsequent use

using hospital or community only controls

Confounders

Adjusted for age, smoking status, BMI, alcohol use, duration of use of oral contraceptives

Conclusions

There is a weak association between third generation oral contraceptives and increased VTE risk compared with second generation oral contraceptives. It is weak enough that a blanket recommendation to choose second generation formulations over third generation ones cannot be made. Instead, the recommendation is for doctors and patients to exercise "clinical prudence," and "informed choice."

Validity

Biases and Flaws

Potential biases

Identified by authors

  • Attrition of susceptibles

Identified by me

  • Recall bias - it is not likely that cases better recall their use of OCPs than controls
  • Overmatching - it is possible that matching based on group general practice registry inadvertently matches on OCP status or the generation of OCP used.
  • Information bias -
  • Exclusion (selection bias) -

Is the study design appropriate to answer the research question?

The study design balances validity of design with practicality, and since VTE is a rare event; case control seems to be an appropriate design.

  • Define "rare event"

Was the sampling scheme and sample size calculation reasonable and appropriate?

Not sure...

Can we generalize from the study population to others, particularly our patient population? Were the measurements valid?

Characteristics of Mount Sinai Patient Population (age, smoking status)

What steps did the researchers take to decrease potential bias?

Researchers tried to:

  • standardize case ascertainment
  • train interviewers
  • audit by re-interview
  • blinded interviewers

Were the statistical tests used appropriate for the research design?

I don't know.

How important are these biases and flaws?

How likely it is to have affected the validity of the results, and figure out in what direction it would affect the results.

Important

Not so important

The Bottom Line

Implications for individual patient cases

A patient is looking into taking hormonal contraceptives in order to decrease menstrual symptoms for a brief period that also coincides with a long transcontinental flight. She is a first-time user.

Implications for practice, policy, training

Not sure

Constituency engagement

Not sure

Next steps in study

  • Prospective cohort, ideally a randomized trial
  • Basic research into fundamentals of progestogen influence on clotting (Recommendation from recent comments in Contraception)

Background Knowledge

Terminology

  • Case-control study - a study design that involves first identifying patients with and without a certain disease (cases and controls), and then interviewing them to determine whether or not or to what extent they were exposed to any number of exposures that could explain or predict the development of disease.
  • Concurrent enrollment - a design choice for case-control studies that involves labeling cases as those diagnosed within a certain time frame. It splits the difference between an incident case (data acquisition occurs when triggered by a new diagnosis) and prevalent case approach (only patients with existing diagnoses at time of data acquisition).
  • Overmatching - it is common to match cases and controls with respect to known confounding exposure variables in order to isolate detectable differences to exposures of interest. However, inadvertent or planned matching on less well-known exposure variables or even the exposure of interest would be a situation termed overmatching.
  • Odds ratios - a measure of association between an exposure variable and an outcome variable (usually development of a disease) that can be computed using data acquired in a case-control study design. The measure is a ratio between the odds of cases having been exposed to the odds of controls having been exposed. The odds for each of the cases and controls, are calculated as the quotient obtained from the division of the number of exposures as dividend by the number of non-exposed as divisor.
  • Logistic regression (unconditional) - a valid regression method that can be applied to analysis of data relating a predictor variable to a categorical outcome, whether binary or ordinal.
  • Conditional logistic regression - an extension of logistic regression that enables one to analyze matched data, and consider continuous and stratified predictor variables, i.e. I have no idea what I'm talking about now.
  • Statistical power - power is the probability that a study/hypothesis test will reject the null hypothesis when the alternative hypothesis is true. It is also dependent on effect size, sample size, valid distribution, and significance threshold (alpha). Larger sample sizes, large effect sizes
  • Bias versus Confounder - confounders are known causal predictors (risk factors) for the outcome of interest that have their own set of correlated variables that can be mistaken for being causally related to the outcome of interest if measured. Bias is a systematic flaw baked into study design or methodology that may influence results one way or the other. Both bias and confounders can be corrected or uncovered by adjusting for certain variables, but it is important to note that a confounder is strictly something that is a known risk factor for the outcome. A bias, for example recall bias, is something that can impact the results, but recalling certain things more vividly is not technically an alternate explanation for an outcome of interest.

Other Links and References

cochrane review 2014