Background: A growing area within medical literature focuses on clinical prediction and decision tools that use variables from patient history, examination, or diagnostic tests to generate outcome-based risk stratification and/or intervention recommendations. Demonstration of appropriate methodologic quality and clinical applicability remains a need. Here we explore tools that advise a complicated, multifaceted phenomenon – the decision to admit a patient to the hospital – in order to assess the basis from which this recommendation is made.
Methods: We searched the MDCalc website (www.mdcalc.com), a free, publicly-available, and widely-used repository to identify calculator tools that provided recommendation for outpatient versus inpatient management. We selected tools designed for adult populations and limited solely to the recommendation for inpatient versus outpatient care, excluding those focused on need for intensive care admission.We used the primary reference listed for each tool on the MDCalc website as the data source for that particular calculator tool. The Wasson-Laupacis framework of methodologic standards for clinical prediction rules was applied to each of the studies. Results were summarized numerically by count and percentage of all studies included. We also assessed whether information necessary to gauge applicability was reported in these studies with respect to their descriptions of patient population, study site, and social determinants of health.
Results: A total of 22 calculators provided hospitalization recommendations for a variety of clinical presentations: chest pain (7), pulmonary embolism (4), community-acquired pneumonia (2), heart failure (2), febrile neutropenia (2), GI bleed (2), syncope (1), TIA (1), and suspected appendicitis (1).All 22 studies reported a clinically significant outcome and 21 of the 22 sufficiently defined this outcome; only 8 studies reported blind assessment of outcome. Predictive variables were identified and defined in 17 studies and blindly assessed in 16 studies. Mathematical techniques were described in all studies, though only 15 provided PPV/NPV or sensitivity/specificity. Only 5 measured the reproducibility of predictive variables and only 1 measured reproducibility of the rule itself. Only 12 studies described any form of prospective validation. No studies prospectively measured the effects of clinical use of the rule.Patient characteristics beyond age, sex, and medical comorbidities were only described in around half of the studies (13/22). Patient race was only included in 2 studies and functional status in 6 studies. Substance use (defined as smoking cigarettes) was described in 7 studies. Mental or behavioral health was never described. All studies indicated the location type and all but 2 specified the geographic area. Eleven studies noted community vs. academic setting; 6 included description of site size; and 5 specified urban, suburban, or rural setting. Only 1 study incorporated any item related to social determinants of health.
Conclusions: When examining the literature underlying clincial risk scoring tools that advise admission or discharge decision-making, we found that methodologic standards are not universally met and information to inform applicability is lacking. These clinical tools focus primarily on specific disease entities and clinical variables which may not encompass the breadth of information necessary to make a disposition determination.
