Clinician-Driven AI: Code-Free Self-Training on Public Data for Diabetic Retinopathy Referral.

Importance: Democratizing artificial intelligence (AI) enables model development by clinicians with a lack of coding expertise, powerful computing resources, and large, well-labeled data sets. Objective: To determine whether resource-constrained clinicians can use self-training via automated machine learning (ML) and public data sets to design high-performing diabetic retinopathy classification models. Design, Setting, and Participants: This diagnostic quality improvement study was conducted from January 1, 2021, to December 31, 2021. A self-training method without coding was used on 2 public data sets with retinal images from patients in France (Messidor-2 [n = 1748]) and the UK and US (EyePACS [n = 58 689]) and externally validated on 1 data set with retinal images from patients of a private Egyptian medical retina clinic (Egypt [n = 210]). An AI model was trained to classify referable diabetic retinopathy as an exemplar use case. Messidor-2 images were assigned adjudicated labels available on Kaggle; 4 images were deemed ungradable and excluded, leaving 1744 images. A total of 300 images randomly selected from the EyePACS data set were independently relabeled by 3 blinded retina specialists using the International Classification of Diabetic Retinopathy protocol for diabetic retinopathy grade and diabetic macular edema presence; 19 images were deemed ungradable, leaving 281 images. Data analysis was performed from February 1 to February 28, 2021. Exposures: Using public data sets, a teacher model was trained with labeled images using supervised learning. Next, the resulting predictions, termed pseudolabels, were used on an unlabeled public data set. Finally, a student model was trained with the existing labeled images and the additional pseudolabeled images. Main Outcomes and Measures: The analyzed metrics for the models included the area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, specificity, and F1 score. The Fisher exact test was performed, and 2-tailed P values were calculated for failure case analysis. Results: For the internal validation data sets, AUROC values for performance ranged from 0.886 to 0.939 for the teacher model and from 0.916 to 0.951 for the student model. For external validation of automated ML model performance, AUROC values and accuracy were 0.964 and 93.3% for the teacher model, 0.950 and 96.7% for the student model, and 0.890 and 94.3% for the manually coded bespoke model, respectively. Conclusions and Relevance: These findings suggest that self-training using automated ML is an effective method to increase both model performance and generalizability while decreasing the need for costly expert labeling. This approach advances the democratization of AI by enabling clinicians without coding expertise or access to large, well-labeled private data sets to develop their own AI models.

The use of biotelemetry to explore disease progression markers in amyotrophic lateral sclerosis.

OBJECTIVE: To explore novel, real-world biotelemetry disease progression markers in patients with amyotrophic lateral sclerosis (ALS) and to compare with clinical gold-standard measures. Methods: This was an exploratory, non-controlled, non-drug 2-phase study comprising a variable length Pilot Phase (n = 5) and a 48-week Core study Phase (n = 25; NCT02447952). Patients with mild or moderate ALS wore biotelemetry sensors for ∼3 days/month at home, measuring physical activity, heart rate variability (HRV), and speech over 48 weeks. These measures were assessed longitudinally in relation to ALS Functional Rating Scale-Revised (ALSFRS-R) score and forced vital capacity (FVC); assessed by telephone [monthly] and clinic visits [every 12 weeks]). Results: Pilot Phase data supported progression into the Core Phase, where a decline in physical activity from baseline followed ALS progression as measured by ALSFRS-R and FVC. Four endpoints showed moderate or strong between-patient correlations with ALSFRS-R total and gross motor domain scores (defined as a correlation coefficient of ≥0.5 or >0.7, respectively): average daytime active; percentage of daytime active; total daytime activity score; total 24-hour activity score. Moderate correlations were observed between speech endpoints and ALSFRS-R bulbar domain scores; HRV data quality was insufficient for reliable assessment. The sensor was generally well tolerated; 6/25 patients reported mostly mild or moderate intensity skin and subcutaneous tissue disorder adverse events. Conclusions: Biotelemetry measures of physical activity in this Pilot Study tracked ALS progression over time, highlighting their potential as endpoints for future clinical trials. A larger, formally powered study is required to further support activity endpoints as novel disease progression markers.

Objectively Monitoring Amyotrophic Lateral Sclerosis Patient Symptoms During Clinical Trials With Sensors: Observational Study.

BACKGROUND: Objective symptom monitoring of patients with Amyotrophic Lateral Sclerosis (ALS) has the potential to provide an important source of information to evaluate the impact of the disease on aspects of real-world functional capacity and activities of daily living in the home setting, providing useful objective outcome measures for clinical trials. OBJECTIVE: This study aimed to investigate the feasibility of a novel digital platform for remote data collection of multiple symptoms-physical activity, heart rate variability (HRV), and digital speech characteristics-in 25 patients with ALS in an observational clinical trial setting to explore the impact of the devices on patients' everyday life and to record tolerability related to the devices and study procedures over 48 weeks. METHODS: In this exploratory, noncontrolled, nondrug study, patients attended a clinical site visit every 3 months to perform activity reference tasks while wearing a sensor, to conduct digital speech tests and for conventional ALS monitoring. In addition, patients wore the sensor in their daily life for approximately 3 days every month for the duration of the study. RESULTS: The amount and quality of digital speech data captured at the clinical sites were as intended, and there were no significant issues. All the home monitoring sensor data available were propagated through the system and were received as expected. However, the amount and quality of physical activity home monitoring data were lower than anticipated. A total of 3 or more days (or partial days) of data were recorded for 65% of protocol time points, with no data collected for 24% of time points. At baseline, 24 of 25 patients provided data, reduced to 13 of 18 patients at Week 48. Lower-than-expected quality HRV data were obtained, likely because of poor contact between the sensor and the skin. In total, 6 of 25 patients had mild or moderate adverse events (AEs) in the skin and subcutaneous tissue disorders category because of skin irritation caused by the electrode patch. There were no reports of serious AEs or deaths. Most patients found the sensor comfortable, with no or minimal impact on daily activities. CONCLUSIONS: The platform can measure physical activity in patients with ALS in their home environment; patients used the equipment successfully, and it was generally well tolerated. The quantity of home monitoring physical activity data was lower than expected, although it was sufficient to allow investigation of novel physical activity end points. Good-quality in-clinic speech data were successfully captured for analysis. Future studies using objective patient monitoring approaches, combined with the most current technological advances, may be useful to elucidate novel digital biomarkers of disease progression.

A randomized, placebo-controlled trial of the analgesic efficacy and safety of the p38 MAP kinase inhibitor, losmapimod, in patients with neuropathic pain from lumbosacral radiculopathy.

OBJECTIVES: Preclinical studies have demonstrated involvement of p38 mitogen-activated protein kinase signaling pathways in the development of persistent pain after peripheral nerve injury. A double-blind, randomized, placebo-controlled study was undertaken to evaluate the analgesic efficacy of losmapimod (GW856553), a novel p38α/ß inhibitor, in patients with chronic neuropathic pain due to lumbosacral radiculopathy. MATERIALS AND METHODS: A total of 144 patients with at least moderate baseline pain intensity (average daily score of ≥4 on an 11-point pain intensity numeric rating scale) were randomized to receive losmapimod, 7.5 mg bid orally or placebo. All patients underwent a blinded placebo run-in period for 7 days before receiving losmapimod/placebo for 28 days. Efficacy and safety evaluations were undertaken weekly. RESULTS: The adjusted mean treatment difference for the change from baseline to week 4 in numeric rating scale was -0.36 U (95% confidence interval, -0.84, 0.13; P=0.149) in favor of losmapimod over placebo; this was not considered clinically meaningful. Statistically significant differences in favor of losmapimod were observed, however, for several secondary endpoints of emotional, physical, and social functioning: Oswestry Disability Index; Profile of Mood States total score; Short-Form 36 Health Survey physical functioning, bodily pain, general health, role emotional, social functioning, and vitality domains; and Short-Form 36 physical, and mental components. There were no unexpected findings related to safety or tolerability following treatment with losmapimod. DISCUSSION: Losmapimod could not be differentiated from placebo in terms of analgesia. The lack of response could reflect insufficient losmapimod levels in the spinal cord or differences between lumbosacral radiculopathy and animal models of neuropathic pain.