Learning Latent Space Representations to Predict Patient Outcomes: Model Development and Validation.

BACKGROUND: Scalable and accurate health outcome prediction using electronic health record (EHR) data has gained much attention in research recently. Previous machine learning models mostly ignore relations between different types of clinical data (ie, laboratory components, International Classification of Diseases codes, and medications). OBJECTIVE: This study aimed to model such relations and build predictive models using the EHR data from intensive care units. We developed innovative neural network models and compared them with the widely used logistic regression model and other state-of-the-art neural network models to predict the patient's mortality using their longitudinal EHR data. METHODS: We built a set of neural network models that we collectively called as long short-term memory (LSTM) outcome prediction using comprehensive feature relations or in short, CLOUT. Our CLOUT models use a correlational neural network model to identify a latent space representation between different types of discrete clinical features during a patient's encounter and integrate the latent representation into an LSTM-based predictive model framework. In addition, we designed an ablation experiment to identify risk factors from our CLOUT models. Using physicians' input as the gold standard, we compared the risk factors identified by both CLOUT and logistic regression models. RESULTS: Experiments on the Medical Information Mart for Intensive Care-III dataset (selected patient population: 7537) show that CLOUT (area under the receiver operating characteristic curve=0.89) has surpassed logistic regression (0.82) and other baseline NN models (<0.86). In addition, physicians' agreement with the CLOUT-derived risk factor rankings was statistically significantly higher than the agreement with the logistic regression model. CONCLUSIONS: Our results support the applicability of CLOUT for real-world clinical use in identifying patients at high risk of mortality.

Patient experience with family history tool: analysis of patients' experience sharing their family health history through patient-computer dialogue in a patient portal.

OBJECTIVE: The collection and use of a family health history are important for assessing the patient's risk of disease, but history taking is often impeded by practical barriers in the office. Provision for patient-computer dialogue, linked with the electronic health record, may enable patients to contribute their history while bypassing these barriers. We sought to assess the patient experience using such a tool. MATERIALS AND METHODS: We linked the family history module of a computer-based medical history to the patient portal of a large academic health system. The interview consisted of 39 primary questions with a predetermined high test-retest reliability. Patients' results were structured and summarized, and available within their electronic health record. Patients optionally completed a survey about their experience. We inductively analyzed free-text responses collected between 2014 and 2016. RESULTS: Among 97 781 patient portal users, 9562 patients accessed and 4223 patients completed the family medical history interview. Of these patients, 1451 completed our survey. Main themes that were identified included (1) patient empowerment, (2) anticipated value, (3) validity concerns, (4) privacy concerns, and (5) reflections on patient-computer dialogue. Patients also provided suggestions for the improvement of future family history tools. DISCUSSION: Patients providing their family health information is an example of collaborative electronic work with clinicians and was seen as valuable by those who participated. Concerns related to contextual information and uncertainty need to be addressed. CONCLUSIONS: Patient-computer dialogue to collect family medical history empowered patients and added perceived value and efficiency to the patient experience of care.

Improving health care proxy documentation using a web-based interview through a patient portal.

OBJECTIVE: Health care proxy (HCP) documentation is suboptimal. To improve rates of proxy selection and documentation, we sought to develop and evaluate a web-based interview to guide patients in their selection, and to capture their choices in their electronic health record (EHR). METHODS: We developed and implemented a HCP interview within the patient portal of a large academic health system. We analyzed the experience, together with demographic and clinical factors, of the first 200 patients who used the portal to complete the interview. We invited users to comment about their experience and analyzed their comments using established qualitative methods. RESULTS: From January 20, 2015 to March 13, 2015, 139 of the 200 patients who completed the interview submitted their HCP information for their clinician to review in the EHR. These patients had a median age of 57 years (Inter Quartile Range (IQR) 45-67) and most were healthy. The 99 patients who did not previously have HCP information in their EHR were more likely to complete and then submit their information than the 101 patients who previously had a proxy in their health record (odds ratio 2.4, P = .005). Qualitative analysis identified several ways in which the portal-based interview reminded, encouraged, and facilitated patients to complete their HCP. CONCLUSIONS: Patients found our online interview convenient and helpful in facilitating selection and documentation of an HCP. Our study demonstrates that a web-based interview to collect and share a patient's HCP information is both feasible and useful.