Clinical prediction models are used frequently in clinical practice to identify patients who are at risk of developing an adverse outcome so that preventive measures can be initiated. A prediction model can be developed in a number of ways; however, an appropriate variable selection strategy needs to be followed in all cases. Our purpose is to introduce readers to the concept of variable selection in prediction modelling, including the importance of variable selection and variable reduction strategies. We will discuss the various variable selection techniques that can be applied during prediction model building (backward elimination, forward selection, stepwise selection and all possible subset selection), and the stopping rule/selection criteria in variable selection (p values, Akaike information criterion, Bayesian information criterion and Mallows’ C_p statistic). This paper focuses on the importance of including appropriate variables, following the proper steps, and adopting the proper methods when selecting variables for prediction models.

Chronic kidney disease (CKD) is a worldwide public health problem, with increasing prevalence, poor outcomes, and high treatment costs.¹ Yet despite the availability of simple laboratory tests to identify people with earlier stages of CKD, there are fewer clinical trials for kidney disease than for other common diseases.² One contributing reason may be that the established end point (ie, a doubling of serum creatinine concentration from baseline, corresponding to a 57% reduction in estimated glomerular filtration rate [GFR]) used to document CKD progression is a late event, requiring long follow-up periods and large sample sizes.^2-4 Improved methods for GFR estimation may allow for use of smaller reductions in estimated GFR (vs a doubling of serum creatinine concentration) as alternative end points to assess CKD progression.^4,5 Evaluation of such alternative end points should include their enumeration and quantification of their relationship with future progression to end-stage renal disease (ESRD) across a wide range of settings. Standardized definitions of CKD progression outcomes would also benefit observational studies and clinical practice.

Medical students carry a large academic load which could potentially contribute to poor sleep quality above and beyond that already experienced by modern society. In this global literature review of the medical students' sleep experience, we find that poor sleep is not only common among medical students, but its prevalence is also higher than in non-medical students and the general population. Several factors including medical students' attitudes, knowledge of sleep, and academic demands have been identified as causative factors, but other potential mechanisms are incompletely understood. A better understanding about the etiology of sleep problems in medical trainees is essential if we hope to improve the overall quality of medical students' lives, including their academic performance. Sleep self-awareness and general knowledge appear insufficient in many studied cohorts, so increasing education for students might be one beneficial intervention. We conclude that there is ample evidence for a high prevalence of the problem, and research in this area should now expand towards initiatives to improve general sleep education for medical students, identify students at risk, and target them with programs to improve sleep.

Background Diabetes mellitus and hypertension are risk factors for acute kidney injury (AKI). Whether estimated glomerular filtration rate (eGFR) and urine albumin-creatinine ratio (ACR) remain risk factors for AKI in the presence and absence of these conditions is uncertain. Study Design Meta-analysis of cohort studies. Setting & Population 8 general-population (1,285,045 participants) and 5 chronic kidney disease (CKD; 79,519 participants) cohorts. Selection Criteria for Studies Cohorts participating in the CKD Prognosis Consortium. Predictors Diabetes and hypertension status, eGFR by the 2009 CKD Epidemiology Collaboration creatinine equation, urine ACR, and interactions. Outcome Hospitalization with AKI, using Cox proportional hazards models to estimate HRs of AKI and random-effects meta-analysis to pool results. Results During a mean follow-up of 4 years, there were 16,480 episodes of AKI in the general-population and 2,087 episodes in the CKD cohorts. Low eGFRs and high ACRs were associated with higher risks of AKI in individuals with or without diabetes and with or without hypertension. When compared to a common reference of eGFR of 80 mL/min/1.73 m² in nondiabetic patients, HRs for AKI were generally higher in diabetic patients at any level of eGFR. The same was true for diabetic patients at all levels of ACR compared with nondiabetic patients. The risk gradient for AKI with lower eGFRs was greater in those without diabetes than with diabetes, but similar with higher ACRs in those without versus with diabetes. Those with hypertension had a higher risk of AKI at eGFRs > 60 mL/min/1.73 m² than those without hypertension. However, risk gradients for AKI with both lower eGFRs and higher ACRs were greater for those without than with hypertension. Limitations AKI identified by diagnostic code. Conclusions Lower eGFRs and higher ACRs are associated with higher risks of AKI among individuals with or without either diabetes or hypertension.