The proportions of patients aged 50 years or older with multiple chronic conditions having surgical procedures are estimated to double in the next 50 years. More than 80 million surgical procedures are performed each year in the United States;these procedures are also increasingly being performed in those with multiple chronic conditions. These patients have the greatest risk of serious complications and procedure-related deaths. Although complication rates from major surgery rise with age and multiple conditions, the effect of specific combinations or sequences of comorbid conditions on outcome is not well understood. Identifying patients at increased risk for complications and adverse outcomes is critical to direct healthcare teams to adjust techniques or interventions, improve decision-making and quality improvement. The most common technique used to quantify the burden of health conditions using secondary data are comorbidity indices, or numeric scores based on a summation of the number of conditions that apply pre-determined "weights" to give certain conditions more importance over others. Comorbidity indices, however, are subject to several limitations and have not been widely incorporated into the routine assessment of patients in clinical care. The availability of a reliable, easy to use and accessible risk prediction tool for adverse events in surgery is essential. We hypothesize that the current techniques for assessing and predicting the relationship of multiple health conditions and outcome may be improved by: 1) using condition specific diagnoses and condition specific outcomes 2) evaluating specific combinations of conditions to assess if their contribution to the risk of outcome is something other than additive and, 3) determining if the temporal sequence of conditions contributes to the prediction of risk beyond the conventional assessment of whether the conditions are present at all. The last decade has seen a rapid increase in the number of available techniques for building predictive models, especially targeting applications with much larger numbers of attributes. This project applies a novel risk prediction strategy to a nationally representative administrative claims database, including longitudinal records from millions of patients. The focus of the project is on the enhanced risk prediction ability of novel dynamic statistical models that will relate the timing, sequence, combination, and clustering of chronic conditions to effectiveness, safety, resource use and cost of the 10 most commonly performed major elective surgical procedures. These novel risk prediction models will utilize dynamic statistical modeling and machine learning techniques to create an easy to use, interactive risk prediction platform. Successfully improving upon a risk prediction tool for adverse events in surgery will better inform patient-centered decision-making, direct healthcare teams to adjust techniques and interventions, help target quality improvement interventions, allow more equitable reimbursement activities and even support accountable care organization activities that rely on accurate estimates of population risk and health.
More than 80 million surgical procedures are performed each year in the United States, increasingly among older patients and those with multiple chronic conditions. Identifying patients at increased risk for complications and adverse outcomes is critical to help inform patient-centered decision-making, direct healthcare teams to adjust techniques and interventions, help target quality improvement interventions, allow more equitable reimbursement activities and even support accountable care organization activities that rely on accurate estimates of population risk and health. Sensitive techniques for risk stratification based on secondary data are necessary, and we propose a novel flexible statistical framework to the use of longitudinal patient data, which defines predictor characteristics for every comorbid condition and relates these to outcome.