Background: Machine learning (ML) techniques improve disease prediction by identifying the most relevant features in multi-dimensional data. We compared the accuracy of ML algorithms for predicting incident diabetic kidney disease (DKD).
Methods: We utilized longitudinal data from 1365 Chinese, Malay and Indian participants aged 40-80 years with diabetes but free of DKD who participated in the baseline and 6-year follow-up visit of the Singapore Epidemiology of Eye Diseases Study (2004-2017). Incident DKD (11.9%) was defined as an estimated glomerular filtration rate (eGFR) <60 mL/min/1.73m2 with at least 25% decrease in eGFR at follow-up from baseline. 339 features including participant characteristics, retinal imaging, genetic and blood metabolites were used as predictors. Performances of several ML models were compared to each other and to logic regression (LR) model based on established features of DKD (age, sex, ethnicity, duration of diabetes, systolic blood pressure, HbA1c, and body mass index) using area under the receiver operating characteristic curve (AUC).
Results: ML model, Elastic Net (EN) had the best AUC (95% confidence interval) of 0.851 (0.847-0.856), which was 7.0% relatively higher than by LR 0.795 (0.790-0.801). Sensitivity and specificity of EN were 88.2% and 65.9% vs. 73.0% and 72.8% by LR. The top-15 predictors included age, ethnicity, antidiabetic medication, hypertension, diabetic retinopathy, systolic blood pressure, HbA1c, eGFR and metabolites related to lipids, lipoproteins, fatty acids and ketone bodies.
Conclusions: Our results showed ML together with feature selection improves prediction accuracy of DKD risk in an asymptomatic stable population and identifies novel risk factors including metabolites.
Funding: This study was supported by the National Medical Research Council, NMRC/OFLCG/001/2017 and NMRC/HCSAINV/MOH-001019-00. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
The authors declare that the data supporting the findings of this study are available within the article and its supplementary materials. The individual participant-level data cannot be shared publicly due to privacy and confidentiality concerns. However, de-identified data may be available upon request with a project proposal to qualified individuals, subject to approval by the Singapore Eye Research Institute and access will require a data sharing agreement. Interested researchers can send data access requests to Prof. Ching-Yu Cheng at the Singapore Eye Research Institute using the following email address: firstname.lastname@example.orgProcessed version of the datasets are provided in Supplementary Tables S1-S5.
- Gavin Tan
- Charumathi Sabanayagam
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Human subjects: SEED was conducted in accordance with the Declaration of Helsinki and was approved by the SingHealth Centralised Institutional Review Board [2018/2717, 2018/ 2921, 2012/487/A, 2015/2279, 2018/2006, 2018/2594, 2018/2570]. Informed consent was obtained from all participants.
- Girish N Nadkarni, Icahn School of Medicine at Mount Sinai, United States
- Received: July 14, 2022
- Accepted: September 12, 2023
- Accepted Manuscript published: September 14, 2023 (version 1)
© 2023, Sabanayagam et al.
This article is distributed under the terms of the Creative Commons Attribution License permitting unrestricted use and redistribution provided that the original author and source are credited.
In most of the world, the mammography screening programmes were paused at the start of the pandemic, whilst mammography screening continued in Denmark. We examined the mammography screening participation during the COVID-19 pandemic in Denmark.
The study population comprised all women aged 50–69 years old invited to participate in mammography screening from 2016 to 2021 in Denmark based on data from the Danish Quality Database for Mammography Screening in combination with population-based registries. Using a generalised linear model, we estimated prevalence ratios (PRs) and 95% confidence intervals (CIs) of mammography screening participation within 90, 180, and 365 d since invitation during the pandemic in comparison with the previous years adjusting for age, year and month of invitation.
The study comprised 1,828,791 invitations among 847,766 women. Before the pandemic, 80.2% of invitations resulted in participation in mammography screening within 90 d, 82.7% within 180 d, and 83.1% within 365 d. At the start of the pandemic, the participation in screening within 90 d was reduced to 69.9% for those invited in pre-lockdown and to 76.5% for those invited in first lockdown. Extending the length of follow-up time to 365 d only a minor overall reduction was observed (PR = 0.94; 95% CI: 0.93–0.95 in pre-lockdown and PR = 0.97; 95% CI: 0.96–0.97 in first lockdown). A lower participation was, however, seen among immigrants and among women with a low income.
The short-term participation in mammography screening was reduced at the start of the pandemic, whilst only a minor reduction in the overall participation was observed with longer follow-up time, indicating that women postponed screening. Some groups of women, nonetheless, had a lower participation, indicating that the social inequity in screening participation was exacerbated during the pandemic.
The study was funded by the Danish Cancer Society Scientific Committee (grant number R321-A17417) and the Danish regions.
Accurate inference of who infected whom in an infectious disease outbreak is critical for the delivery of effective infection prevention and control. The increased resolution of pathogen whole-genome sequencing has significantly improved our ability to infer transmission events. Despite this, transmission inference often remains limited by the lack of genomic variation between the source case and infected contacts. Although within-host genetic diversity is common among a wide variety of pathogens, conventional whole-genome sequencing phylogenetic approaches exclusively use consensus sequences, which consider only the most prevalent nucleotide at each position and therefore fail to capture low frequency variation within samples. We hypothesized that including within-sample variation in a phylogenetic model would help to identify who infected whom in instances in which this was previously impossible. Using whole-genome sequences from SARS-CoV-2 multi-institutional outbreaks as an example, we show how within-sample diversity is partially maintained among repeated serial samples from the same host, it can transmitted between those cases with known epidemiological links, and how this improves phylogenetic inference and our understanding of who infected whom. Our technique is applicable to other infectious diseases and has immediate clinical utility in infection prevention and control.