KM Research Hub
← 返回论文库
已完成 入排:排除 图源:可得 入排置信度 0.98

Development of a Novel Machine Learning Method for Estimation of Life-Long Chronic Disease Progression and Its Application to Type 2 Diabetes.

Yamato Sano, Ryota Jin, Hideki Yoshioka, Yuki Nakazato, Hiromi Sato, Akihiro Hisaka

Clin Transl Sci 18(10) e70351

PMID41127917
PMCIDPMC12547360
DOI10.1111/cts.70351
发表日期2025-10-23
入排依据:未发现恶性肿瘤人群证据。
摘要

摘要

Individual predictions of long-term chronic disease progression from data of limited duration provide valuable insights into estimating patient outcomes and therapeutic needs. Statistical Restoration of Fragmented Time course (SReFT) was developed to address this challenge, yet it is computationally too intensive for large-scale datasets. Although diabetes is a representative chronic disease with significant medical needs, it has been challenging to analyze long-term changes using large-scale patient data due to this limitation. In this study, we aimed to develop a new method (SReFT-machine learning, SReFT-ML) by applying machine learning to the concept of SReFT, and to confirm its performance using synthetic data and the data from a clinical trial, the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial (N = 10,251). SReFT-ML has successfully analyzed both synthetic and clinical data, and reconstructed biomarker trajectories over a 30-year period in patients with diabetes. Decreases in diastolic blood pressure and renal function may be important indicators of disease progression. Furthermore, although age and mortality data were not included in the model, survival analysis demonstrated a clear trend of hazard increases in mortality and diabetes-related outcomes with disease progression. This study introduced machine learning to enhance long-term disease progression modeling. The resulting model characterized a 30-year trajectory of disease risk in diabetes. The results provide a clinically meaningful hypothesis that incorporating systemic factors such as renal function and blood pressure, in addition to classic glycemic control, may enhance comprehensive diabetes care. Trial Registration: ClinicalTrials.gov number: NCT00000620.

KM 提取结果

1 个 Figure · 2 个 KM 面板

进入结果中心
KM Figure2 个面板

FIGURE 5

FIGURE 5 Survival analysis for mortality against disease time. (a) Survival function estimated from Kaplan–Meier estimator for all‐cause mortality. (b) Hazard function estimated from Nelson–Aalen estimator for all‐cause mortality. (c) Survival function estimated from Kaplan–Meier estimator for cardiovascular event‐related mortality. (d) Hazard function estimated from Nelson–Aalen estimator for cardiovascular event‐related mortality. Blue lines indicate survival functions (a, c) and hazard functions (b, d), and fill indicates 95% CI. The orange line represents the estimated Gompertz distribution. The hazard function was smoothed using the “Epanechnikov kernel” function. CI, confidence intervals.

A

all-cause mortality

FIGURE 5 · Panel A

complete 待 EXE 对齐 预期 2 条 已提取 2 条

Panel A 是同一 FIGURE 5 中的独立 KM 子图;人群、终点或分析条件以本 Panel 图注语义为准。

查看结果
C

cardiovascular event-related mortality

FIGURE 5 · Panel C

complete 待 EXE 对齐 预期 2 条 已提取 2 条

Panel C 是同一 FIGURE 5 中的独立 KM 子图;人群、终点或分析条件以本 Panel 图注语义为准。

查看结果
其他 Figure(5,默认折叠)

这些图片未产生 KM 曲线结果,不在主页面重复展示。

FIGURE 1FIGURE 1 Schematic overview of SReFT and SReFT‐ML. Both SReFT and SReFT‐ML aim to model long‐term disease progression by analyzing changes in biomarkers over time. The primary objective of both methods is to predict disease time and understand the progression of chronic diseases using fragmented clinical data. While SReFT uses a nonlinear mixed‐effects model to estimate disease time and biomarker changes, SReFT‐ML leverages machine learning techniques to achieve the same goals. One of the key differences between the two methods is their computational efficiency. SReFT‐ML is designed to be more computationally efficient and scalable, making it suitable for analyzing large‐scale datasets with numerous biomarkers and patients. ML, machine learning; SReFT, Statistical Restoration of Fragmented Time course.
非 KM Figure
FIGURE 2FIGURE 2 Concept of disease time estimation (a) and network structure of SReFT‐ML (b). SReFT‐ML employs two neural networks to estimate disease time and biomarker changes, which allows it to handle large datasets and complex data structures without the need for initial parameter values. In panel (a), Network 1 aims to estimate the order of t offset which is the disease time at study entry and uses the intercept ( x intercept ) and slope ( x slope ) of fragments which were obtained from a prior linear mixed‐effects model analysis. Network 2 uses all the observed biomarker values and creates the model so that the observation match its prediction, thereby also adjusting t offset . In panel (b), red circles indicate the first neural network, and blue circles indicate the second neural network. Light, normal, and dark colors indicate input layer, hidden layer, and output layer. The number of units was adjusted according to the size of the feature. The gray lines indicate the weights. Subscript j denotes the biomarkers. ML, machine learning; SReFT, Statistical Restoration of Fragmented Time course; t obs , observation time; t offset , offset time.
非 KM Figure
FIGURE 3FIGURE 3 SReFT‐ML analysis using synthetic data. (a) The solid lines represent estimated curves, and the dashed lines represent curves with the set parameters. Color coding corresponds to groups of covariates. Each gray point represents individual observations. (b) Horizontal axis represents the synthesized t offset , and vertical axis represents the t offset estimated by SReFT‐ML. (c) Results of survival analysis on simulation. Please note that age and mortality data were not utilized in the SReFT‐ML analysis, and survival analysis results were presented to confirm the consistency of the analysis. Blue lines indicate survival function estimated from Kaplan–Meier estimator (left panel) and hazard function estimated from Nelson–Allen estimators (right panel), and fill indicate 95% CI. The orange line shows the estimated Gompertz distribution. ACCORD, Action to Control Cardiovascular Risk in Diabetes; CI, confidence intervals; ML, machine learning; SReFT, Statistical Restoration of Fragmented Time course; t offset , offset time.
非 KM Figure
FIGURE 4FIGURE 4 SReFT‐ML analysis using data from the ACCORD trial. The solid lines represent estimated curves for each biomarker. Color coding corresponds to groups of covariates (i.e., glycemic treatment groups). Each gray point represents an individual observation. ALT, alanine aminotransferase; brazindex, SF36 Aggregate Interference Score; CPK, creatine phosphokinase; DBP, diastolic blood pressure; DSST, DANTES Subject Standardized Tests Total Score; DTSQ, Diabetes Treatment Satisfaction Questionnaire; eGFR, estimated glomerular filtration rate from 4 variable Modification of Diet in Renal Disease (MDRD) equation; FPG, fasting plasma glucose; HDL, high density lipoprotein; HR, heart rate; HUI2pf, Health Utilities Index Mark 2 (HUI2); HUI3Scor, Health Utilities Index Mark 3 (HUI3); K, potassium; LDL, low density lipoprotein; Mental, SF36 Aggregate Mental Score; MMSE, Mini‐Mental State Examination Total Score; Neuropathy, Michigan Neuropathy Screening Instrument score; PHQ9, PHQ (Patient Health Questions) Depression Scale; Physical, SF36 Aggregate Physical activity Score; RAVLT, Rey Auditory Verbal Learning Test Score; SBP, systolic blood pressure; SES, Summarized Symptom and Distress Score; SReFT, Statistical Restoration of Fragmented Time course; STROOP, Stroop Interference Score; TC, total cholesterol; TG, triglyceride; UALB, urinary albumin; UCRE, urinary creatinine; VLDL, very low density lipoprotein.
非 KM Figure
FIGURE 6FIGURE 6 Comparison of microvascular outcomes in diabetic neuropathy and diabetic eye complications among groups stratified by t offset ‐quartiles. Hazard ratios were calculated using proportional hazards regression models. The statistical test for each outcome was derived from a model that accounted for the groups stratified by t offset ‐quartiles, using the first quartile as the reference. CI, confidence intervals; HR, hazard ratio; MNSI, Michigan neuropathy screening instrument; Q2, second quartile of t offset ; Q3, third quartile of t offset ; Q4, fourth quartile of t offset ; t offset , offset time.
非 KM Figure