Benchmarking can include comparing an ICU's actual performance with predicted performance. The increased interoperability of medical devices, electronic health records (EHRs) and information systems has improved the acquisition and presentation of data to healthcare professionals. This data has enabled the training of predictive models. However, thi s plethora of data sources has also introduced new risks that societal bias will lead to machine learning systems with fairness issues for patient groups. In addition, when variations in data documentation are non-random, significant bias can be introduced, improving, or worsening measured performance for an institution relative to peers. This work focuses on ICU mortality benchmarking. In particular, we analyze the fairness of a model based on Generalised Additiv e Models (GAM) [ 3 ] that predicts mortality in the ICU. This model is used to compare actual versus predicted outcom es to assess ICU performance.

Complex deep learning models show high prediction tasks in various clinical prediction tasks but their inherent complexity makes it more challenging to explain model predictions for clinicians and healthcare providers. Existing research on explainability of deep learning models in healthcare have two major limitations: using post-hoc explanations and using raw clinical variables as units of explanation, both of which are often difficult for human interpretation. In this work, we designed a self-explaining deep learning framework using the expert-knowledge driven clinical concepts or intermediate features as units of explanation. The self-explaining nature of our proposed model comes from generating both explanations and predictions within the same architectural framework via joint training. We tested our proposed approach on a publicly available Electronic Health Records (EHR) dataset for predicting patient mortality in the ICU. In order to analyze the performance-interpretability trade-off, we compared our proposed model with a baseline having the same set-up but without the explanation components. Experimental results suggest that adding explainability components to a deep learning framework does not impact prediction performance and the explanations generated by the model can provide insights to the clinicians to understand the possible reasons behind patient mortality.

Trauma mortality results from a multitude of non-linear dependent risk factors including patient demographics, injury characteristics, medical care provided, and characteristics of medical facilities; yet traditional approach attempted to capture these relationships using rigid regression models. We hypothesized that a transfer learning based machine learning algorithm could deeply understand a trauma patient's condition and accurately identify individuals at high risk for mortality without relying on restrictive regression model criteria. Anonymous patient visit data were obtained from years 2007-2014 of the National Trauma Data Bank. Patients with incomplete vitals, unknown outcome, or missing demographics data were excluded. All patient visits occurred in U.S. hospitals, and of the 2,007,485 encounters that were retrospectively examined, 8,198 resulted in mortality (0.4%). The machine intelligence model was evaluated on its sensitivity, specificity, positive and negative predictive value, and Matthews Correlation Coefficient. Our model achieved similar performance in age-specific comparison models and generalized well when applied to all ages simultaneously. While testing for confounding factors, we discovered that excluding fall-related injuries boosted performance for adult trauma patients; however, it reduced performance for children. The machine intelligence model described here demonstrates similar performance to contemporary machine intelligence models without requiring restrictive regression model criteria or extensive medical expertise.

The study, published in PLOS One, adds to the growing evidence that AI could revolutionise healthcare in the UK and beyond. So far, the emphasis has been on the potential of AI to help diagnose and treat various diseases, but these new findings suggest it could also help predict the likelihood of patients dying too. "It won't be long before doctors are routinely using these sorts of tools in the clinic to make better diagnoses and prognoses, which can help them decide the best ways to care for their patients," says Crick scientist Andrew Steele, first author of the paper. "Doctors already use computer-based tools to work out whether a patient is at risk of heart disease, and machine-learning will allow more accurate models to be developed for a wider range of conditions." The model was designed using the electronic health data of over 80,000 patients, collected as part of routine care, and available for researchers on the CALIBER platform.

By using an artificially intelligent algorithm to predict patient mortality, a research team from Stanford University is hoping to improve the timing of end-of-life care for critically ill patients. In tests, the system proved eerily accurate, correctly predicting mortality outcomes in 90 percent of cases. But while the system is able to predict when a patient might die, it still cannot tell doctors how it came to its conclusion. Doctors must consider an array of complex factors, ranging from a patient's age and family history to their response to drugs and the nature of the affliction itself. To complicate matters, doctors have to contend with their own egos, biases, or an unconscious reluctance to assess a patient's prospects for what they are.

ICU mortality scoring systems attempt to predict patient mortality using predictive models with various clinical predictors. Examples of such systems are APACHE, SAPS and MPM. However, most such scoring systems do not actively look for and include interaction terms, despite physicians intuitively taking such interactions into account when making a diagnosis. One barrier to including such terms in predictive models is the difficulty of using most variable selection methods in high-dimensional datasets. A genetic algorithm framework for variable selection with logistic regression models is used to search for two-way interaction terms in a clinical dataset of adult ICU patients, with separate models being built for each category of diagnosis upon admittance to the ICU. The models had good discrimination across all categories, with a weighted average AUC of 0.84 (>0.90 for several categories) and the genetic algorithm was able to find several significant interaction terms, which may be able to provide greater insight into mortality prediction for health practitioners. The GA selected models had improved performance against stepwise selection and random forest models, and provides greater flexibility in terms of variable selection by being able to optimize over any modeler-defined model performance metric instead of a specific variable importance metric.