Application of Machine Learning Techniques to Identify Data Reliability and Factors Affecting Outcome After Stroke Using Electronic Administrative Records

Santu Rana, Wei Luo, Truyen Tran, Svetha Venkatesh, Paul Talman, Thanh G. Phan, Dinh Phung, Benjamin ClissoldFrontiers in Neurology 2021 · DOI

Introduction

Predicting a patient's discharge destination following a stroke is a critical component of hospital resource management and clinical planning. While specialized clinical registries are the traditional "gold standard" for building predictive models, they are often resource-intensive to maintain and may lack the breadth of data found in general hospital records. This study explores whether "messy" electronic administrative records (EHR)—data typically collected for billing and management—can be transformed into highly accurate predictive tools using machine learning.

Methodology

The researchers employed penalized logistic regression with Lasso (Least Absolute Shrinkage and Selection Operator). Lasso is particularly well-suited for high-dimensional administrative data because it performs automatic feature selection, effectively "zeroing out" irrelevant variables from a large pool of candidates.

The study compared two primary feature sets:

  1. Administrative Model: Included 1,303 features derived from ICD-10 diagnostic codes, procedure codes, and demographic data.
  2. Baseline Clinical Model: Included a smaller set of "gold standard" stroke-specific factors like age, smoking status, hypertension, and imaging types.

The models were validated using a temporal split, training on data from 2009–2012 and testing on a subsequent cohort from 2012–2013 to ensure robustness.

Patient Population

The study utilized a dataset of 2,531 patients from Barwon Health, a regional tertiary hospital in Victoria, Australia. The cohort represented a wide range of ages and stroke subtypes.

Demographic and clinical characteristics of the 2,531 patient cohort, detailing age groups, stroke types, and comorbidities.

As shown in the patient characteristics, Ischemic stroke was the most frequent diagnosis (37.7%), followed by Transient Ischemic Attack (TIA, 25.1%) and Intracerebral Hemorrhage (ICH, 14.6%). The population was notably elderly, with over 40% of patients aged 80 or older.

Results

Predictive Performance

The administrative EHR model consistently provided better outcome predictions than the model limited to specific clinical factors. The performance was measured using the Area Under the Curve (AUC):

  • TIA: Achieved an impressive AUC of 0.948 for predicting discharge to home vs. other destinations.
  • Ischemic Stroke: Predicted the need for a nursing home vs. death with an AUC of 0.902.
  • ICH: Predicted discharge to rehabilitation vs. other outcomes with an AUC of 0.850.

Identification of Risk Factors

Because administrative data lacks direct functional scores (like the NIHSS), the Lasso model identified powerful "proxy markers" for patient health and stroke severity.

Lasso regression coefficients identifying key factors influencing stroke outcomes. Positive values favor better outcomes, while negative values indicate risk factors for poorer outcomes.

As illustrated in the coefficient plot, factors such as advanced age (80-90 and >90), prior ventilatory support, and urinary incontinence (R32) were strong negative predictors. Conversely, the need for specific diagnostic imaging like Magnetic Resonance Angiography (MRA) or MRI Brain often correlated with more favorable outcomes, likely indicating patients who were clinically stable enough to undergo advanced diagnostics.

Discussion and Conclusion

The findings suggest that the "noise" in administrative data actually contains valuable signals about a patient's overall frailty and functional status. While administrative data relies on the accuracy of hospital coders and lacks granular clinical nuance (such as infarct volume), its breadth allows for a comprehensive assessment of risk.

In conclusion, electronic administrative records produce reliable outcome predictions and identify clinically appropriate factors impacting recovery. This approach offers a cost-effective means of identifying modifiable factors associated with patient discharge, potentially aiding in clinical decision-making and improving patient education regarding expected outcomes.

Summary updated

Abstract

<b>Aim:</b> To use available electronic administrative records to identify data reliability, predict discharge destination, and identify risk factors associated with specific outcomes following hospital admission with stroke, compared to stroke specific clinical factors, using machine learning techniques. <b>Method:</b> The study included 2,531 patients having at least one admission with a confirmed diagnosis of stroke, collected from a regional hospital in Australia within 2009-2013. Using machine learning (penalized regression with Lasso) techniques, patients having their index admission between June 2009 and July 2012 were used to derive predictive models, and patients having their index admission between July 2012 and June 2013 were used for validation. Three different stroke types [intracerebral hemorrhage (ICH), ischemic stroke, transient ischemic attack (TIA)] were considered and five different comparison outcome settings were considered. Our electronic administrative record based predictive model was compared with a predictive model composed of "baseline" clinical features, more specific for stroke, such as age, gender, smoking habits, co-morbidities (high cholesterol, hypertension, atrial fibrillation, and ischemic heart disease), types of imaging done (CT scan, MRI, etc.), and occurrence of in-hospital pneumonia. Risk factors associated with likelihood of negative outcomes were identified. <b>Results:</b> The data was highly reliable at predicting discharge to rehabilitation and all other outcomes vs. death for ICH (AUC 0.85 and 0.825, respectively), all discharge outcomes except home vs. rehabilitation for ischemic stroke, and discharge home vs. others and home vs. rehabilitation for TIA (AUC 0.948 and 0.873, respectively). Electronic health record data appeared to provide improved prediction of outcomes over stroke specific clinical factors from the machine learning models. Common risk factors associated with a negative impact on expected outcomes appeared clinically intuitive, and included older age groups, prior ventilatory support, urinary incontinence, need for imaging, and need for allied health input. <b>Conclusion:</b> Electronic administrative records from this cohort produced reliable outcome prediction and identified clinically appropriate factors negatively impacting most outcome variables following hospital admission with stroke. This presents a means of future identification of modifiable factors associated with patient discharge destination. This may potentially aid in patient selection for certain interventions and aid in better patient and clinician education regarding expected discharge outcomes.
BibTeX
@inproceedings{rana2021application,
  author = {Santu Rana and Wei Luo and Truyen Tran and Svetha Venkatesh and Paul Talman and Thanh G. Phan and Dinh Phung and Benjamin Clissold},
  title = {{Application of Machine Learning Techniques to Identify Data Reliability and Factors Affecting Outcome After Stroke Using Electronic Administrative Records}},
  booktitle = {Frontiers in Neurology},
  year = {2021},
  doi = {10.3389/fneur.2021.670379},
}
Figure