Elsevier

Injury

Volume 41, Issue 8, August 2010, Pages 869-873
Injury

Comparison of artificial neural network and logistic regression models for predicting mortality in elderly patients with hip fracture

https://doi.org/10.1016/j.injury.2010.04.023Get rights and content

Abstract

Purpose

Older patients with hip fracture have a mortality rate one year after surgery of 20–30%. The purpose of this study is to establish a predictive model to assess the outcome of surgical treatment in older patients with hip fracture.

Methods

A database of information from 286 consecutive cases of surgery for hip fracture from the Department of Orthopedics, National Taiwan University Hospital Yun-Lin Branch, was utilised for model building and testing. Both logistic regression and artificial neural network (ANN) models were developed. Cases were randomly assigned to training and testing datasets. A testing dataset was utilised to test the accuracy of both models (n = 89).

Results

The areas under the receiver operator characteristic curves of both models were utilised to compare predictability and accuracy. The logistic regression training and testing datasets had an area of 0.938 (95% CI: 0.904, 0.972) and 0.784 (95% CI: 0.669, 0.899), respectively, below the 0.998 (95% CI: 0.995, 1.000) and 0.949 (95% CI: 0.857, 1.000) of the final ANN model.

Conclusion

Overall, ANNs have higher predictive ability than logistic regression, perhaps because they are not affected by interactions between factors. They may assist in complex decision making in the clinical setting.

Introduction

The most common fractures in the elderly are hip fractures, spinal compression fractures and distal radial fractures. Of these, hip fractures result in the highest mortality, around 20–30% per year, as well as the worst prognosis. Only 50% of patients recover pre-fracture mobility.6, 7 The incidence rate of hip fractures in the elderly has doubled in the past 20 years and is expected to double again by the year 2050.2 In Taiwan, the National Health Insurance Program inpatient database shows an incidence rate of hip fractures (1996–2000) of 225 per 100,000 in males and 505 per 100,000 in females, about 6 times that reported for Beijing, 1.2 times higher that of Hong Kong and on par with US Caucasian rates. Prevalence amongst older males in Taiwan is slightly higher than for US males.6, 27 Optimal treatment of hip fractures focuses on aggregate, surgical method, bone density, operating time points and other variables affecting mobility and mortality.15, 18, 23, 28, 31

Outcome prediction studies have proven useful in many areas of health care research, especially in critical care and trauma. For a predictive model to be useful it must be simple to calculate, generalisable and have an apparent structure.33 Amongst the methods used for outcome prediction, artificial neural networks (ANNs) excel at pattern recognition. Artificial neural networks are mathematical constructs that use previously solved examples to build a system of neurons to make new decisions, classify and forecast.12, 30 ANN models have been applied in diagnosing myocardial infarction,18, 28 pulmonary emboli23 and gastrointestinal haemorrhage and conditions.12, 33 Nillson et al. used ANNs to select risk variables and predict mortality after cardiac surgery using data from 18,362 patients. The neural networks selected 34 relevant risk factors to predict mortality and results were superior to those produced by logistic regression models.21

Few published reports have comprehensively assessed the impact of patient health and operation methodology on outcomes of surgical treatment in patients with hip fractures. The purpose of this study is to establish a model using ANN to predict one-year mortality after surgical treatment of elderly patients with hip fracture. The goal of such model is to help older patients and their doctors better assess the risks involved in surgery, and do the proper decision making.

Section snippets

Patients and study design

A total of 286 older adults (>65 years) with hip fracture were enrolled from January 2005 to May 2007. Data was obtained from the Department of Orthopedics, National Taiwan University Hospital Yun-Lin Branch. Inclusion criteria were patients (1) having undergone surgery for hip fracture, (2) older than 65 years old, (3) low energy trauma fractures and (4) follow-up for >1 year or having died within one year of surgery. Exclusion criteria were: (1) pathological fracture, (2) ineligible for

Patient data results and modelling

The database was divided randomly into two sets: 197 cases (70%) for training and 89 (30%) for model testing. Only the training dataset was used to build models. Characteristics of the training and testing dataset are summarised in Table 1. The two datasets did not differ significantly for any of the 12 variables studied. Before modelling, we boosted the training dataset to balance the number of survival and non-survival cases.

Modelling

The binary variable of death or survival was the output variable of

Discussion

Hip fracture in older adults is complicated by the many variables affecting treatment outcomes in older patients, including preexisting comorbidities, presence of functional limitations and hospital type. Many studies had pointed out that mortality after hip fracture operation in the elderly correlates with functional recovery.5, 11, 19 Even more, the literature indicates that physicians’ personal experience correlates with the outcome of surgical treatment. Smektala et al.28 found that

Conflict of interest statement

No competing interests exist.

Acknowledgements

The authors thank the National Taiwan University Hospital Yun-Lin Branch and National Yunlin University of Science and Technology for subjects and laboratory equipment.

References (33)

  • M.J.A. Berry et al.

    Data mining technique for marketing, sale, and customer support

    (1997)
  • S. Boonen et al.

    Functional outcome and quality of life following hip fracture in elderly women: a prospective controlled study

    Osteoporos Int

    (2004)
  • W.C. Chie et al.

    High incidence rate of hip fracture in Taiwan: estimated from a nationwide health insurance database

    Osteoporos Int

    (2004)
  • B. Eftekhar et al.

    Comparison of artificial neural network and logistic regression models for prediction of mortality in head trauma based on initial clinical data

    BMC Med Inform Decis Making

    (2005)
  • E. Golec et al.

    Remote results and failures operating treatment pertrochanteric femur fractures using dynamic hip stabilizer

    Chir Narzadow Ruchu Ortop Pol

    (2007)
  • D. Greenwood

    An overview of neural networks

    Behav Sci

    (1991)
  • Cited by (53)

    • Which criteria is a better predictor of ICU admission in trauma patients? An artificial neural network approach

      2022, Surgeon
      Citation Excerpt :

      ANN are new computing methods for ML, displaying knowledge, and applying knowledge to the vast majority of output responses from complex systems.29 Limited studies have been conducted to predict the likelihood of mortality, duration of stay in the ICU, and other health-related outcomes using the ANN model and compare it with LR.30–34 Importantly, studies in traumatic patients often focus on the main outcome, including early and late mortality.

    • Machine Learning Model Developed to Aid in Patient Selection for Outpatient Total Joint Arthroplasty

      2022, Arthroplasty Today
      Citation Excerpt :

      However, ML models act as a “black box” because only the input predictive variables and output predictions are known and relationships and strengths of individual variables remain unknown [14]. Although ANN models have better predictive ability, multivariate logistic regression is useful to determine which individual factors are associated with the outcomes [11]. The ANN models in this study used variables found on multiple linear regression to be significantly associated with total LOS and variables found on multivariate logistic regression analyses to be significantly associated with same-day discharge.

    View all citing articles on Scopus
    View full text