Reporting and methods in clinical prediction research: a systematic review

Area of Science:

• Medical Research Methodology
• Clinical Prediction Modeling

Background:

• Assesses the reporting and methodology of prediction studies.
• Focuses on aims, designs, participant selection, outcomes, predictors, statistical power, methods, and performance measures.

Purpose of the Study:

• To evaluate adherence to current methodological recommendations in published prediction studies.
• To identify common deficiencies in the conduct and reporting of prediction research.

Main Methods:

• Conducted a hand search of prediction studies published in 2008 in six high-impact general medical journals.
• Developed a systematic scoring system based on recent recommendations for prediction research.
• Two reviewers independently scored 71 retrieved papers.

Main Results:

• Study designs were unclear in 15% of studies; prospective cohorts were most common (60%).
• Continuous predictors were dichotomized in 32% of studies, contrary to recommendations.
• Statistical power (events per predictor) was inadequate in 53% of studies where determinable.
• Predictive performance (calibration, discrimination) was reported in only 12% and 27% of studies, respectively.
• Many studies (29%) used p-value cut-offs for predictor selection.

Conclusions:

• The majority of prediction studies in high-impact journals do not adhere to current methodological standards.
• These deficiencies limit the reliability and applicability of findings from prediction research.
• Improvements in study conduct and reporting are needed to enhance the value of prediction models.