Systematic variable reduction for simplification of incisional hernia risk prediction instruments

• Published in: The American journal of surgery Vol. 224; no. 1; pp. 576 - 583
• Main Authors: McAuliffe, Phoebe B., Hsu, Jesse Y., Broach, Robyn B., Borovskiy, Yuliya, Christopher, Adrienne N., Morris, Martin P., Fischer, John P.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.07.2022; Elsevier Limited
• Subjects: Abdomen; Abdominal surgery; Chronic obstructive pulmonary disease; Clinical outcomes; Codes; Collagen; Colorectal surgery; Electronic health records; Gender; Gynecology; Hernia; Hernias; Incisional hernia; Laparotomy; Medical records; Model accuracy; Obstetrics; Paralysis; Patients; Point of care tool; Population; Prediction models; Risk prediction; Social history; Statistical models; Surgical outcomes; Variable reduction; Variables; Abdominal surgery; Variable reduction; Point of care tool; Risk prediction; Laparotomy; Incisional hernia
• ISSN: 0002-9610; 1879-1883
• DOI: 10.1016/j.amjsurg.2022.03.003

Incisional hernia (IH) is a complex, costly and difficult to manage surgical complication. We aim to create an accurate and parsimonious model to assess IH risk, pared down for practicality and translation in the clinical environment. Institutional abdominal surgical patients from 2002 to 2019 were identified (N = 102,281); primary outcome of IH, demographic factors, and comorbidities were extracted. A 32-variable Cox proportional hazards model was generated. Reduced-variable models were created by systematic removal of variables 1–4 and 23–25 at a time. The c-statistic of the full 32-variable model was 0.7232. Four comorbidities decreased accuracy of the model: COPD, paralysis, cancer and combined autoimmune/hereditary collagenopathy or AAA diagnosis. The model with those 4 comorbidities removed had the highest c-statistic (0.7291). The most reduced model included 7 variables and had a c-statistic of 0.7127. Accuracy of an IH predictive model is only marginally affected by a vast reduction in end-user inputs. •Incisional hernia is a highly morbid surgical complication.•Predictive models aim to assess risk of incisional hernia after laparotomy.•Complex predictive models can involve a multitude of variables.•Variable-reduced predictive models are nearly as predictive as full-scale models.•Parsimonious models integrate into workflow to make a real-world impact.