QUESTION 1:
What was the main research question of this study? (1 sentence)
QUESTION 2:
How (including the source) and when (including the index date) were the cases defined? (1-2 sentences)
QUESTION 3:
What specific type of bias may have affected the selection of cases as defined above? Please justify your answer. (2-3 sentences)
QUESTION 4:
The authors used a risk-set sampling strategy (also known as incidence density sampling) to extract sex and age matched controls from the Danish Person Registry. This means that cases and controls were also matched on length of person-time follow-up, thereby allowing for the calculation of measures of association between exposure and disease (in this case, odds ratios) that can be interpreted as an estimate of the incidence rate ratio (RR). Explain why this method of selecting controls guaranteed a proper selection of controls. (2-3 sentences)
QUESTION 5:
If some of the controls were in fact cases, how may this have affected the magnitude of the association (ORs) reported in this study? Please justify your answer. (1-2 sentences)
QUESTION 6:
How was the main exposure of interest (antibiotic use) defined and measured in this study? (1-2 sentences)
QUESTION 7:
The way that authors collected the information on the exposure meant that they avoided two key types of measurement error/bias which commonly affect case-control studies. Name one of these types of measurement bias and in your own words, describe what this bias is? (1-2 sentences)
QUESTION 8:
Levels of the exposure were quantified according to the number of antibiotic courses given before the index date. Specifically, subjects were allocated into 3 categories of exposure: 0-1 antibiotic courses (the reference category), 2-4 antibiotic courses and ≥5 antibiotic courses (data details shown in Table 1). Crude (or unadjusted) and adjusted association estimates (ORs) were then calculated for the 2-4 (vs. 0-1) antibiotic courses and for ≥5 (vs. 0.1) antibiotic courses (shown in Table 2)
QUESTION 8a:
The authors did not report on the levels of compliance to the prescribed antibiotic treatment (i.e. whether study participants actually took the antibiotics that they were prescribed).
Assuming that there was some measurement error in the classification of exposure categories, and it was likely a similar degree of misclassification regardless of whether participants were cases or controls, i) what is the likely impact that this would have on the magnitude of the ORs reported in the present study? ii) would this be an example of differential or non- differential misclassification - please justify your answer. (2-3 sentences)
QUESTION 8b:
Suppose you actually had data on compliance to prescribed antibiotic treatments. You discover that 30% of cases and 30% of controls that were originally categorised as having been prescribed 2-4 courses of antibiotics only actually took 0-1 courses in total. Using the data from Table 1, recalculate the new odds ratio for the association between actual number of courses of antibiotics taken and type 2 diabetes for those actually taking 2-4 courses compared to those taking 0-1 courses of antibiotics. Please show your new 2x2 table and calculations to receive full marks.
QUESTION 8c:
The authors chose to create a single category (0-1 antibiotic courses) to merge those with no antibiotic use (0 antibiotic courses) together with those who had received 1 previous antibiotic course to create the reference category. 15,809 of the cases and 180,653 of the controls did not receive any antibiotic courses. Using this information together with relevant information from Table 1, calculate the crude odds ratio (OR) for 1 vs. 0 antibiotic courses. Note: To receive full marks, please display the 2x2 table and full calculations.
QUESTION 8d:
How do you interpret the value of the OR obtained in question 8c? (1 sentence)
QUESTION 8e:
Based on your answers to questions 8c and 8d above, do you think that merging the individuals with 0 and with 1 antibiotic courses into one category for analysis is appropriate? Please justify your answer (2-3 sentences)
QUESTION 9:
In this study the authors assessed and considered the potential effects of several confounders in their analyses.
QUESTION 9a:
Which confounders were accounted for in this study, and how was this done? (2 sentences)
QUESTION 9b:
What was the extent and direction of the confounding effects of age? (1 sentence)
QUESTION 9c:
Examining the results from Table 2, describe and compare the net effects of adjustment for confounding for 2-4 courses of antibiotics and ≥5 courses of antibiotics. In your comparison, also discuss how you assessed the impact of adjustment for confounders. (3-4 sentences)
QUESTION 9d:
An important type of confounding occurring in pharmacoepidemiological studies (like this study) is that of confounding by indication (acknowledged by the authors). Explain, in your own words, the meaning of this type of confounding within the context of the study by Mikkelsen et al., 2015. (3-4 sentences)
QUESTION 9e:
Do you think that the authors accounted for all confounders relevant for the association under investigation? Please justify your answer. (3-4 sentences)
QUESTION 10:
This study did not include information about the gut microbiota composition of cases and controls, the mechanisms through which exposure to antibiotics is thought to lead to type 2 diabetes. If a measure of gut microbiota had been obtained as part of the study, how would you classify such a variable: as a potential confounder, mediator or effect-modifier? Please justify your answer (2-3 sentences)
QUESTION 11:
The authors mentioned that "the observed associations between exposure and type 2 diabetes were identical when stratified by age, gender and observation periods (data not shown)".
The results of these stratified analyses thus ruled out effect-modification by these variables. Explain, in your own words, the meaning of effect modification. (2-3 sentences)
QUESTION 12:
Describe the level of precision of the effect estimates presented in Table 3 for different groups of antibiotics. How did you assess this? What do you think is the main driver of differences in precision for different groups of antibiotics? (3-4 sentences)
QUESTION 13:
Do you consider the findings of this study to be generalizable to other populations? Please justify your answer. (4-6 sentences)
QUESTION 14:
Do the authors attempt to explore a dose-response relationship in this study? If yes, please justify your answer with examples from the paper; if no, please provide examples of how the authors may have explored a dose-response relationship (2-3 sentences)
QUESTION 15:
Of all Bradford-Hill's criteria for causality, the criterion of temporality is the only one that is essential to demonstrate causality. What is meant by temporality? Do you think this study has established a temporal relationship? Please justify. (3-4 sentences).
QUESTION 16:
In the discussion section of the paper the authors mentioned that another study (by Boursi et al., The effect of past antibiotic exposure on diabetes risk. Eur J Endocrinol, 2015) showed ORs for diabetes risk associated with antibiotic use similar to the associations observed in this study. Which of the Bradford Hill's criteria does this statement offer support for? (1 sentence)
Paper: Mikkelsen KH, Knop FK, Frost M, Hallas J, Pottegard A. Use of antibiotics and risk of type 2 diabetes: A population-based case-control study. The Journal of Clinical Endocrinology and Metabolism. 2015; 100(10):3633-3640