Selection bias
    1a. Percentage loss to follow-up? Only applicable to cohort studies
        >20% or unknown or unclassifiable0The percentage is unclassifiable if the total eligible cohort is not clear.
    1b. Percentage response of the cases and controls? Only applicable to case-control studies
        <75% or unknown or unclassifiable0Response = [1 − (refusal of subject/physician, contact problems, death before interview) / eligible subjects) × 100. Judging the response of cases and controls, the lowest percentage counts. Unclassifiable if the complete eligible group is not clear.
    2. Was the absolute difference in percentage response <20% between cases and controls?
        Not applicable (cohort study)10
        No or unknown0
    3. Percentage incident cases?
        <75%0Some studies include prevalent and/or fatal cases without information about the date of diagnosis, which may introduce bias, as physical activity may also be associated with survival.
        75-99% or unknown4
    4. Did the cases and controls originate from the same source population?
        Not applicable (cohort study)10
        No or unknown0
    5. Were the same exclusion/inclusion criteria applied to cases and controls?
        Not applicable (cohort study)7
        No or unknown0
    Maximal selection bias score42
Misclassification bias
    Determination of physical activity
        6. Was the measure of leisure time activities that was analyzed complete enough?
            No0Activity may have been assessed extensively whereas only few components where included in the analysis. Judgement of the item was based on reviewers' consensus.
        7. Was total activity assessed?
            No0Total activity means leisure time activity and job/household activity. These should be combined in one effect measure.
        8. Did the measure of physical activity include intensity, frequency and duration?
            One component or unknown0
            Intensity + frequency/duration per week5
        9. Type of administration of physical activity questionnaire
            By proxy0By proxy means that physical activity is not individually assessed (e.g., a family member was asked); classification was based on college registration of athletics.
        10. Was the operationalization of the physical activity score understandable?
            No or partly0
        11. Did the physical activity measure include past physical activity?
            No, only recent0A physical activity measure covering more periods of life is supposed to be more accurate.
            Yes, more life periods4
        12. Did the authors consider changes over time in physical activity pattern in the analyses?
            No0Yes: e.g., when two measurements of physical activity several years apart were used to classify participants in consistently active or inactive.
        13. Was the physical activity questionnaire validated or was reliability tested?
            No or unknown0
        14. Was physical activity level assessed before endometrial cancer diagnosis?
            No0When physical activity is measured after the diagnosis of endometrial cancer, there is a possibility of recall bias.
            No, but physical activity level was assessed the same way for cases and controls4
        15. Was the case diagnosis valid?
            No or unknown0Endometrial cancer self-report may be valid, especially if confirmed by medical report.
        16. Could benign endometrial disease (carcinoma in situ) in any way have influenced the results?
            Yes or unknown0Any influence seems unlikely if <5% of cases have carcinoma in situ and/or if separate analyses/exclusion did not result in different estimates.
    Maximal misclassification bias score42
Confounding bias
    17. Were confounders adjusted for in a correct way (statistically)?
        No or unknown0
    18. Could residual confounding be a problem?
        Yes0Potential confounders: age, BMI, parity, age at menopause (and/or smoking), oral contraceptive use, hormone replacement therapy; Residual confounding could also be a problem when (a) continuous variables were crudely categorized or (b) BMI or hormone replacement therapy were not measured within 5 y of diagnosis.
    19. Were the effects of leisure time activities adjusted for occupational/household activities?
        No or unknown0
    Maximal confounding bias score21
Maximal total score: 105
    Patients and controls are not from the same source populations when:
    •hospital-based controls are used and it is unlikely that the controls would be referred to the same hospital in case of cancer;
    •cases are recruited from a specialized cancer hospital (e.g., subgroup of patients with more advanced disease) and controls are population based and it is quite uncertain whether they would be referred to the same hospital if they had become a case;
    •controls are not a random sample of the source population;
    •cases and controls are selected from different areas/countries.