The Evidence Speaks Series is a recurring feature highlighting the latest in Advancing Health research. This series features summaries of select publications and is designed to keep media and the research community up to date with the Centre’s current research results in the health outcomes field.
To ensure this research is quick and easy to share, you are welcome to save the social cards and use as you see fit.
Predictors of mortality for patients who leave hospitals early against medical advice
Predicting drug overdose and death after “before medically advised” hospital discharge. Naik H, Daly-Grafstein D, Hu X, Khan M, Kaasa BM, Brubacher JR, Nasmith T, Lyden JR, Moe J, Crabtree A, Slaunwhite AK, Staples JA. CMAJ. 2025.
When a patient leaves the hospital before the doctors think they’re ready, it’s called a Before Medically Advised (BMA) discharge. BMAs present a complex challenge for clinicians, hospitals, and patients, as patients who initiate a BMA discharge are about twice as likely to die and about 10 times more likely to experience an illicit drug overdose compared to patients undergoing routine, physician-advised hospital discharge. Specific interventions like low-barrier follow-ups, opioid agonist therapy (OAT), and estimates of an individual’s risk of adverse outcomes post- BMA discharge could help reduce those outcomes. A team of researchers including Advancing Health scientists, Drs. Amanda Slaunwhite, and John Staples and postdoctoral fellow Dr. aimed to develop prediction models to estimate the risk of death from any cause in the first 30 days among all patients who initiated a BMA discharge and the risk of illicit overdose in the first 30 days for patients with a history of substance use.
The research was done with two cohorts to differentiate between patients with regular health concerns and those with substance use disorders: Death from any cause (excluding Cohort A) with 6,440 patients and illicit drug overdose (Cohort B) with 4,466 patients. They found that in Hiten Naik Cohort A, 102 deaths (1.6 per cent) occurred within the first 30 days of BMA discharge. Modelling based on Cohort A showed that strong predictors of death in this group include comorbidities (multiple health issues), a history of heart disease, and a history of cancer. In Cohort B, 223 (5.2 per cent) illicit drug overdoses occurred within 30 days of BMA discharge. Modelling revealed that the strong predictors of drug overdose after BMA discharge in Cohort B included homelessness, receipt of social income assistance, a history of opioid use disorder, a history of non-alcohol substance use disorder, at least one drug overdose in the past year, and discharge from a surgery. The predictive models developed by the team serve as a starting point for identifying patients who are high risk and could benefit from more support. Future research should explore whether these adverse outcomes for BMA discharge patients are not just predictable but also preventable, and how they can be prevented.
Risk factors for complications in traumatic spinal cord injury: A retrospective analysis of a cohort of patients identified from administrative data. Bond M, Beresford A, Noonan V, Rotem-Kohavi N, Fallah N, Dvorak M, Kwon B, Liu G, Sutherland J. J. Spinal Cord Med. 2025.
When severe damage is dealt to a person’s spinal cord, this can cause a traumatic spinal cord injury (TSCI). Because the spinal cord is responsible for sending neural signals to different parts of the body, a TSCI can be devastating, causing major changes in sensation, movement, strength, and other bodily functions. In order to help address these negative impacts, researchers wanted to identify other factors associated with the severity of TSCI symptoms.
Advancing Health scientist, Dr. Jason Sutherland and team analysed hospital records that were linked with administrative databases for 3,344 patients TSCI patients in BC from 2001 to 2021. They looked at in-hospital mortality rate, length of stay in acute care, and other negative side-effects. Their analysis found that patients with more negative outcomes tended to be elderly, have other illnesses or conditions (comorbidities), and also have more severe initial injuries, including traumatic brain injuries. The researchers suggest that patients with these characteristics should be identified early and carefully monitored to prevent unnecessary complications. Future research should investigate treatments and policies to better support these patients.
Machine learning using clinical variables to screen for depression and anxiety in people with early multiple sclerosis. Phillips B, Morrow SA, Singh H, Oh J, Kolind S, Lynd LD, Prat A, Tam R, Traboulsee A, Patten SB, Group CS. Mult Scler Relat Disord. 2025.
Depression and anxiety are more prevalent in people with multiple sclerosis (MS) than in the general population, making psychiatric comorbidities screening an important component of MS care. But since large-scale screening efforts that use scales to measure depressive symptoms and identify a subgroup with a high probability of depression have not been successful, machine learning algorithms that use routinely collected clinical information could serve the same function more efficiently.
To that end, Advancing Health scientist Dr. Larry Lynd and a team of researchers conducted a study that aimed to utilize demographic and clinical measures from baseline date (collected from a pan-Canadian cohort of people with MS). They then developed models with machine learning that can predict patient scores on depression and anxiety questionnaires. By splitting the data into training sets that coached the algorithm and testing sets that evaluated model performance, the team found the machine learning model performed well for depression but was less useful in predicting anxiety. They also analysed which features of anxiety and depression were most predictive of risks. The researchers found that self-reported psychosocial fatigue was strongly predictive of both depression and anxiety, as was self-reported cognitive fatigue. A greater number of comorbidities predicted a greater risk for both depression and anxiety. The findings of this study indicate that it is possible to screen for depression and, to a lesser extent, anxiety in people with MS without needing full self-report questionnaires and complex scales, making testing more time-efficient for clinicians and patients. Because this study was only able to develop a successful model for anxiety, future research should explore machine learning methods like deep learning that could produce better predictors of anxiety and other clinical outcomes.
