- Identify all available breast cancer risk prediction models that include family history of breast cancer as a predictor.
- Qualitatively summarise the characteristics of these risk prediction models.
- Appraise the predictive accuracy (validation, discrimination and calibration) of these models.
- Compare model performance by meta-analysing model performance statistics across studies, if feasible.
- Appraise the predictive accuracy (validation, discrimination and calibration) of models validated (or developed) in ‘elevated risk’ women (i.e. those at increased risk of developing breast cancer based on their family history). We anticipate that relevant studies may include (a) participants enrolled in a screening programme; (b) those with at least one first-degree relative with breast cancer; (c) those attending a family risk service.
- Compare model performance in ‘elevated risk’ women by meta-analysing model performance statistics across studies on model development or validation in ‘elevated risk’ women if feasible.
- Make recommendations for breast cancer risk prediction models that may be useful for patient management in the breast cancer family risk setting.
The review question has been outlined according to the PICOTS system (Debray 2017) in Table 1.Investigation of sources of heterogeneity between studies
As with all meta-analysis it is important to understand the cause of between-study heterogeneity in model performance. It is important to investigate potential sources of heterogeneity in discrimination and calibration across validation studies, to understand when the model performance remains adequate and when it may require further improvements (Debray 2017). Heterogeneity may be due to differences in the study (e.g. in terms of design, follow-up time, or outcome definition), differences in the statistical analysis or characteristics related to selective reporting and publication (e.g. risk of bias, study size) and/or populations e.g. case-mix or baseline risk differences. We anticipate that the most relevant sources of heterogeneity between studies will be differences in study population/case-mix. From scoping the literature it is apparent that diversity exists in populations used in breast cancer risk prediction model development and validation; for example, populations differed in terms of ethnicity, geographical location and baseline risk category, all of which could contribute to differences in model performance between validation studies. To a lesser extent, we anticipate that outcome definition may be a source of heterogeneity between studies with some studies defining breast cancer as all breast cancer (invasive and ductal carcinoma in situ (DCIS)) and others potentially defining breast cancer as invasive breast cancer only. These sources of heterogeneity will be investigated and reported in the review.
This is a protocol.