Clinical Deployment of Machine Learning Tools in Transplant Medicine: What Does the Future Hold?

Medical applications of machine learning (ML) have shown promise in analyzing patient data to support clinical decision-making and provide patient-specific outcomes. In transplantation, several applications of ML exist which include pretransplant: patient prioritization, donor-recipient matching, organ allocation, and posttransplant outcomes. Numerous studies have shown the development and utility of ML models, which have the potential to augment transplant medicine. Despite increasing efforts to develop robust ML models for clinical use, very few of these tools are deployed in the healthcare setting. Here, we summarize the current applications of ML in transplant and discuss a potential clinical deployment framework using examples in organ transplantation. We identified that creating an interdisciplinary team, curating a reliable dataset, addressing the barriers to implementation, and understanding current clinical evaluation models could help in deploying ML models into the transplant clinic setting.

Glucagon-like peptide-1 receptor agonists and sodium-glucose cotransporter-2 inhibitors for the treatment of diabetes mellitus in liver transplant recipients.

AIM: To investigate the efficacy and safety of glucagon-like peptide-1 receptor agonists (GLP-1RAs) and sodium-glucose cotransporter-2 (SGLT2) inhibitors in liver transplant (LT) recipients with diabetes. METHODS: A single-centre, retrospective analysis of prospectively collected data from an LT recipient database (1990-2023) was conducted. We included adults with pre-existing diabetes and post-transplant diabetes, newly started on GLP-1RAs and/or SGLT2 inhibitors after LT. Metabolic and biochemical parameters and outcomes were collected for up to 12 months after starting medications and were compared to those in patients receiving dipeptidyl peptidase-4 (DPP-4) inhibitors. Statistical analysis included descriptive statistics and linear mixed models. RESULTS: We included participants on GLP-1RAs (n = 46), SGLT2 inhibitors (n = 87), combination therapy (n = 12), and a DPP-4 inhibitor comparator (n = 217). Both GLP-1RAs and combination therapy decreased mean glycated haemoglobin (HbA1c) levels, and combination therapy remained significant when adjusted for DPP-4 inhibitor treatment (-3.5%, 95% CI [-6.1, -0.95]; p = 0.0089) at 12 months. All three groups had significant decreases in mean weight and body mass index, but these remained significant in the GLP-1RA (-5.2 kg, 95% CI [-8.7, -1.7], p = 0.0039 and 1.99 kg/m2, 95% CI [-3.4, -0.6], p = 0.0048) and combination therapy groups (-5.4 kg, 95% CI [-10.5, -0.36], p = 0.04 and -3.4 kg/m2, 95% CI [-5.5, -1.3], p = 0.0015) when adjusted for DPP-4 inhibitor treatment at 12 months. Alanine aminotransferase levels decreased with GLP-1RA and combination therapy. There were two (1.4%) cases of graft rejection. CONCLUSION: We found that GLP-1RAs, SGLT2 inhibitors, and their combination, led to significant weight loss in LT recipients with diabetes. Combination therapy, in particular, lowered HbA1c and alanine aminotransferase levels compared to DPP-4 inhibitors. Further studies are needed to assess long-term safety and efficacy.

Harnessing Metabolites as Serum Biomarkers for Liver Graft Pathology Prediction Using Machine Learning.

Graft injury affects over 50% of liver transplant (LT) recipients, but non-invasive biomarkers to diagnose and guide treatment are currently limited. We aimed to develop a biomarker of graft injury by integrating serum metabolomic profiles with clinical variables. Serum from 55 LT recipients with biopsy confirmed metabolic dysfunction-associated steatohepatitis (MASH), T-cell mediated rejection (TCMR) and biliary complications was collected and processed using a combination of LC-MS/MS assay. The metabolomic profiles were integrated with clinical information using a multi-class Machine Learning (ML) classifier. The model's efficacy was assessed through the Out-of-Bag (OOB) error estimate evaluation. Our ML model yielded an overall accuracy of 79.66% with an OOB estimate of the error rate at 19.75%. The model exhibited a maximum ability to distinguish MASH, with an OOB error estimate of 7.4% compared to 22.2% for biliary and 29.6% for TCMR. The metabolites serine and serotonin emerged as the topmost predictors. When predicting binary outcomes using three models: Biliary (biliary vs. rest), MASH (MASH vs. rest) and TCMR (TCMR vs. rest); the AUCs were 0.882, 0.972 and 0.896, respectively. Our ML tool integrating serum metabolites with clinical variables shows promise as a non-invasive, multi-class serum biomarker of graft pathology.

Clinical Deployment of Machine Learning Tools in Transplant Medicine: What Does the Future Hold?

Medical applications of machine learning (ML) have shown promise in analyzing patient data to support clinical decision-making and provide patient-specific outcomes. In transplantation, several applications of ML exist which include pretransplant: patient prioritization, donor-recipient matching, organ allocation, and posttransplant outcomes. Numerous studies have shown the development and utility of ML models, which have the potential to augment transplant medicine. Despite increasing efforts to develop robust ML models for clinical use, very few of these tools are deployed in the healthcare setting. Here, we summarize the current applications of ML in transplant and discuss a potential clinical deployment framework using examples in organ transplantation. We identified that creating an interdisciplinary team, curating a reliable dataset, addressing the barriers to implementation, and understanding current clinical evaluation models could help in deploying ML models into the transplant clinic setting.

A deep learning framework for personalised dynamic diagnosis of graft fibrosis after liver transplantation: a retrospective, single Canadian centre, longitudinal study.

BACKGROUND: Recurrent graft fibrosis after liver transplantation can threaten both graft and patient survival. Therefore, early detection of fibrosis is essential to avoid disease progression and the need for retransplantation. Non-invasive blood-based biomarkers of fibrosis are limited by moderate accuracy and high cost. We aimed to evaluate the accuracy of machine learning algorithms in detecting graft fibrosis using longitudinal clinical and laboratory data. METHODS: In this retrospective, longitudinal study, we trained machine learning algorithms, including our novel weighted long short-term memory (LSTM) model, to predict the risk of significant fibrosis using follow-up data from 1893 adults who had a liver transplantation between Feb 1, 1987, and Dec 30, 2019, with at least one liver biopsy post transplantation. Liver biopsy samples with indefinitive fibrosis stage and those from patients with multiple transplantations were excluded. Longitudinal clinical variables were collected from transplantation to the date of last available liver biopsy. Deep learning models were trained on 70% of the patients as the training set and 30% of the patients as the test set. The algorithms were also separately tested on longitudinal data from patients in a subgroup of patients (n=149) who had transient elastography within 1 year before or after the date of liver biopsy. Weighted LSTM model performance for diagnosing significant fibrosis was compared against LSTM, other deep learning models (recurrent neural network and temporal convolutional network), and machine learning models (Random Forest, Support vector machines, Logistic regression, Lasso regression, and Ridge regression) and aspartate aminotransferase-to-platelet ratio index (APRI), fibrosis-4 index (FIB-4), and transient elastography. FINDINGS: 1893 people who had a liver transplantation (1261 [67%] men and 632 [33%] women) with at least one liver biopsy between Jan 1, 1992, and June 30, 2020, were included in the study (591 [31%] cases and 1302 [69%] controls). The median age at liver transplantation was 53·7 years (IQR 47·3-59·0) for cases and 55·3 years (48·0 to 61·2) for controls. The median time interval between transplant and liver biopsy was 21 months (5 to 71). The weighted LSTM model (area under the curve 0·798 [95% CI 0·790 to 0·810]) consistently outperformed other methods, including unweighted LSTM (0·761 [0·750 to 0·769]; p=0·031) Recurrent Neural Network (0·736 [0·721 to 0·744]), Temporal Convolutional Networks (0·700 [0·662 to 0·747], and Random Forest 0·679 [0·652 to 0·707]), FIB-4 (0·650 [0·636 to 0·663]) and APRI (0·682 [0·671 to 0·694]) when diagnosing F2 or worse stage fibrosis. In a subgroup of patients with transient elastography results, weighted LSTM was not significantly better at detecting fibrosis (≥F2; 0·705 [0·687 to 0·724]) than transient elastography (0·685 [0·662 to 0·704]). The top ten variables predictive for significant fibrosis were recipient age, primary indication for transplantation, donor age, and longitudinal data for creatinine, alanine aminotransferase, aspartate aminotransferase, total bilirubin, platelets, white blood cell count, and weight. INTERPRETATION: Deep learning algorithms, particularly weighted LSTM, outperform other routinely used non-invasive modalities and could help with the earlier diagnosis of graft fibrosis using longitudinal clinical and laboratory variables. The list of most important predictive variables for the development of fibrosis will enable clinicians to modify their management accordingly to prevent onset of graft cirrhosis. FUNDING: Canadian Institute of Health Research, American Society of Transplantation, Toronto General and Western Hospital Foundation, and Paladin Labs.

Axial Impairment Following Deep Brain Stimulation in Parkinson's Disease: A Surgicogenomic Approach.

BACKGROUND: Postoperative outcome following deep brain stimulation (DBS) of the subthalamic nucleus is variable, particularly with respect to axial motor improvement. We hypothesized a genetic underpinning to the response to surgical intervention, termed "surgicogenomics". OBJECTIVE: We aimed to identify genetic variants associated with clinical heterogeneity in DBS outcome of Parkinson's disease (PD) patients that could then be applied clinically to target selection leading to improved surgical outcome. METHODS: Retrospective clinical data was extracted from 150 patient's charts. Each individual was genotyped using the genome-wide NeuroX array tailored to study neurologic diseases. Genetic data were clustered based on surgical outcome assessed by comparing pre- and post-operative scores of levodopa equivalent daily dose and axial impairment at one and five years post-surgery. Allele frequencies were compared between patients with excellent vs. moderate/poor outcomes grouped using a priori defined cut-offs. We analyzed common variants, burden of rare coding variants, and PD polygenic risk score. RESULTS: NeuroX identified 2,917 polymorphic markers at 113 genes mapped to known PD loci. The gene-burden analyses of 202 rare nonsynonymous variants suggested a nominal association of axial impairment with 14 genes (most consistent with CRHR1, IP6K2, and PRSS3). The strongest association with surgical outcome was detected between a reduction in levodopa equivalent daily dose and common variations tagging two linkage disequilibrium blocks with SH3GL2. CONCLUSION: Once validated in independent populations, our findings may be implemented to improve patient selection for DBS in PD.

Effects of Two Training Programs on Transcriptional Levels of Neurotrophins and Glial Cells Population in Hippocampus of Experimental Multiple Sclerosis.

The aim of the present study was to investigate the effects of high-intensity interval training (HIIT) versus low-intensity continuous training (LICT) on transcriptional levels of neurotrophic factors and oligodendrocyte/microglia cell loss in a cuprizone (CP) induced animal model of demyelination. Male C57BL/6 mice were assigned to six groups: control (C), cuprizone-induced demyelination (CP), interval training (IT), continuous training (CT), IT plus CP (ITCP), and CT plus CP (CTCP). Training programs on the treadmill were performed for four weeks, and then demyelination was induced by feeding mice a diet containing 0.2% cuprizone for five weeks. Animals that received cuprizone showed poorer motor function, lower expression of BDNF, GDNF, NGF, and fewer oligodendrocytes in the hippocampus compared to the control animals. The numbers of oligodendrocyte and microglia cells increased in the ITCP group compared to the CTCP group (P<0.05). Both training programs increased the mRNA levels of BDNF, GDNF and NGF, and the HIIT program was more effective than the LICT program (P<0.05). Both exercise programs prevented the abnormal neurological movements induced by cuprizone. Our results indicated that HIIT versus LICT had a greater neuroprotective effect against multiple sclerosis by improving gene expression for abnormal neurotrophins and cellular loss in the hippocampus.

Channelopathy-related SCN10A gene variants predict cerebellar dysfunction in multiple sclerosis.

OBJECTIVE: To determine the motor-behavioral and neural correlates of putative functional common variants in the sodium-channel NaV1.8 encoding gene (SCN10A) in vivo in patients with multiple sclerosis (MS). METHODS: We recruited 161 patients with relapsing-onset MS and 94 demographically comparable healthy participants. All patients with MS underwent structural MRI and clinical examinations (Expanded Disability Status Scale [EDSS] and Multiple Sclerosis Functional Composite [MSFC]). Whole-brain voxel-wise and cerebellar volumetry were performed to assess differences in regional brain volumes between genotype groups. Resting-state fMRI was acquired from 62 patients with MS to evaluate differences in cerebellar functional connectivity. All participants were genotyped for 4 potentially functional SCN10A polymorphisms. RESULTS: Two SCN10A polymorphisms in high linkage disequilibrium (r(2) = 0.95) showed significant association with MSFC performance in patients with MS (rs6795970: p = 6.2 × 10(-4); rs6801957: p = 0.0025). Patients with MS with rs6795970(AA) genotype performed significantly worse than rs6795970(G) carriers in MSFC (p = 1.8 × 10(-4)) and all of its subscores. This association was independent of EDSS and cerebellar atrophy. Although the genotype groups showed no difference in regional brain volumes, rs6795970(AA) carriers demonstrated significantly diminished cerebellar functional connectivity with the thalami and midbrain. No significant SCN10A-genotype effect was observed on MSFC performance in healthy participants. CONCLUSIONS: Our data suggest that SCN10A variation substantially influences functional status, including prominent effects on motor coordination in patients with MS. These findings were supported by the effects of this variant on a neural system important for motor coordination, namely cerebello-thalamic circuitry. Overall, our findings add to the emerging evidence that suggests that sodium channel NaV1.8 could serve as a target for future drug-based interventions to treat cerebellar dysfunction in MS.

Postinjury treatment with magnesium sulfate attenuates neuropathic pains following spinal cord injury in male rats.

Spinal cord injury (SCI) has a number of severe and disabling consequences including chronic pain. Approximately 40% of patients experience neuropathic pain, which appears to be persistent. Previous studies have demonstrated the neuroprotective effects of magnesium sulfate (MgSO4). We aimed to investigate the effect of MgSO4 on neuropathic pains following SCI in male rats. Thirty-two adult male rats (weight 300-350 g) were used. After laminectomy, a complete SCI was induced by compression of the spinal cord for 1 min with an aneurysm clip. A single dose of 300 or 600 mg/kg MgSO4 was injected intraperitoneally. Tail-flick latency and acetone drop test scores were evaluated before surgery and once a week for 4 weeks after surgery. Rats in groups SCI+Mg300 and SCI+Mg600 showed significantly higher mean tail-flick latencies and lower mean scores in the acetone test compared with those in the SCI+veh group 4 weeks after surgery (P<0.05). These findings revealed that systemic single-dose administration of MgSO4 can attenuate thermal hyperalgesia and cold allodynia induced by SCI in rats.