A Neural Network Analysis of the Effect of High and Low Frailty Index Indicators On Predicting Elective Surgery Discharge Destinations

Steven Walczak ,Vic Velanovich 



Frailty is frequently used by clinicians to help determine surgical outcomes. The frailty index, which represents the frequency of frailty indicators present in an individual, is one method for evaluating patient frailty to predict surgical outcomes. However, the frailty index treats all indicators of frailty that are used in the index as equivalent. Our hypothesis is that frailty indicators may be divided into groups of high and low-impact indicators and this separation will improve surgical discharge outcome prediction accuracy.


Prediction of surgical discharge outcomes prior to an operation is beneficial to clinicians, post-operative care providers, patients, and their families [1]. Therefore, it is important to only use independent variables in the prediction model that are available prior to surgery to enable effective post-operative care planning, improved recovery, and stress reduction for patients.
The frailty index (FI) has been shown to be an effective indicator of surgical patient outcomes [2, 3], especially with elderly patients [4], or patients whose health is otherwise compromised affecting their ability to cope with stressors [5]. The accumulating deficits FI model of Rockwood and Mitnitski [6, 7], which represents frailty as a single value composed of the sum of all deficits present in a patient divided by the total number of deficits measured, initially recommended having 40 or more deficit variables. Prior research however has shown that a much smaller collection of variables is able to produce reliable FI and these smaller variable models are referred to as modified FI (mFI) [8–10].

Materials and methods

This study used anonymized historic data from a large national database freely available to member hospitals participating in the American College of Surgeons National Surgical Quality Improvement Program (NSQIP). Due to the use of retrospective anonymized data this study was given a waiver for need for consent from the USF Office of Research Integrity and Compliance.
All data is acquired from the American College of Surgeons National Surgical Quality Improvement Program (NSQIP) for the 2018 Participant Use Data (PUF) files, which has 1020511 available records reported from 722 hospitals across 49 states in the United States and 11 other countries [21]. The data is constrained to only use elective inpatient operations to provide a more homogenous patient group as trauma and emergency surgery patients have categorically different preoperative physiological insults and often have ongoing disabilities following surgery which affect discharge destination [22]. Using only inpatient elective operations reduces the total number of NSQIP surgical records available to 390185. The only exclusion criterion is those records missing values for the corresponding independent variables for the ANN model.


The best performing ANN architectures for each specialty and corresponding mFI configuration, are given in the S2 Appendix. Results of the 20 ANN discharge destination prediction models, utilizing either a single mFI or split high-impact and low-impact mFIs is given in Table 2, with percent improvement of the split high-impact and low-impact mFIs over the traditional model calculated as:


The reported research has demonstrated that for most types of operation, the splitting of an mFI into two mFI (one for high-impact indicators and one for low-impact indicators) used in an ANN model where different weights are applied to each of the separate mFI indicator types produces improved accuracy for destination discharge outcome predictions. This directly supports our H0 research hypothesis for eight out of the nine surgical specialties studied and for the composite set of all surgeries. The cardiac surgical specialty results contradict H0. If H0 is rewritten as 10 new hypotheses H1 to H10 including the surgical specialty, such as: Surgical outcome prediction models that treat high-impact frailty indicators differently from low-impact frailty indicators will increase outcome prediction accuracy for general surgery, then 9 of the 10 hypotheses are confirmed and only 1 is rejected.


While frailty has been shown to be a reliable indicator of surgical outcomes in prior research, the equal treatment of frailty indicators implies that every health deficit is identical regarding the ability for a patient to recovery from surgery. Our research demonstrates using ANN models, that frailty factors need to be divided into two groups: high-impact indicators and low-impact indicators. This knowledge discovery in medicine will facilitate improvement of clinical informatics systems for post-surgical prognostic planning and care which in turn will improve the health outcomes of elective surgery patients.

Citation: Walczak S, Velanovich V (2023) A neural network analysis of the effect of high and low frailty index indicators on predicting elective surgery discharge destinations. PLoS ONE 18(4): e0284206.

Editor: Zyad James Carr, Yale School of Medicine: Yale University School of Medicine, UNITED STATES

Received: December 13, 2022; Accepted: March 24, 2023; Published: April 7, 2023

Copyright: © 2023 Walczak, Velanovich. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All relevant data are within the manuscript and its Supporting Information files.

Funding: The author(s) received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.

Harvard Medical School - Leadership in Medicine Southeast Asia47th IHF World Hospital CongressHealthcare CNO Summit - USAHealthcare CMO Summit - USA