Developing and validating a clinlabomics-based machine-learning model for early detection of retinal detachment in patients with high myopia

Abstract

Background: Retinal detachment (RD) is a vision-threatening disorder of significant severity. Individuals with high myopia (HM) face a 2 to 6 times higher risk of developing RD compared to non-myopes. The timely identification of high myopia-related retinal detachment (HMRD) is crucial for effective treatment and prevention of additional vision impairment. Consequently, our objective was to streamline and validate a machine-learning model based on clinical laboratory omics (clinlabomics) for the early detection of RD in HM patients.

Methods: We extracted clinlabomics data from the electronic health records for 24,440 HM and 5607 HMRD between 2015 and 2022. Lasso regression analysis assessed fifty-nine variables, excluding collinear variables (variance inflation factor > 10). Four models based on random forest, gradient boosting machine (GBM), generalized linear model, and Deep Learning Model were trained for HMRD diagnosis and employed for internal validation. An external test of the models was done. Three random data sets were further processed to validate the performance of the diagnostic model. The primary outcomes were the area under the receiver operating characteristic curve (AUC) and the area under the precision-recall curve (AUCPR) to diagnose HMRD.

Results: Nine variables were selected by all models. Given the AUC and AUCPR values across the different sets, the GBM model was chosen as the final diagnostic model. The GBM model had an AUC of 0.8550 (95%CI = 0.8322-0.8967) and an AUCPR of 0.5584 (95%CI = 0.5250-0.5879) in the training set. The AUC and AUCPR in the internal validation were 0.8405 (95%CI = 0.8060-0.8966) and 0.5355 (95%CI = 0.4988-0.5732). During the external test evaluation, it reached an AUC of 0.7579 (95%CI = 0.7340-0.7840) and an AUCPR of 0.5587 (95%CI = 0.5345-0.5880). A similar discriminative capacity was observed in the three random data sets. The GBM model was well-calibrated across all the sets. The GBM-RD model was implemented into a web application that provides risk prediction for HM individuals.

Conclusion: GBM algorithms based on nine features successfully predicted the diagnosis of RD in patients with HM, which will help ophthalmologists to establish a preliminary diagnosis and to improve diagnostic accuracy in the clinic.

Keywords: Clinlabomics; Detection; High myopia; Machine-learning; Retinal detachment.

Fig. 1

Study flow-chart: This figure displays the participant flow-chart. GLU: glucose; PCT: thrombocytocrit; MPV: mean platelet volume; UA: uric acid; APTT: activated partial thromboplastin time; GLB: globulin; BASP: percentage of basophil. GBM: gradient boosting machine; GLM: generalized linear model

Fig. 2

The top 20 significant variables chosen by five machine learning models (A-E) and the intersection set of these variables (F). TBA: total bile acid; TBIL: total bilirubin; TC: total cholesterol; N: neutrophil number; ALB: albumin; RBC: red blood count; HBA1C: glycosylated hemoglobin; PCT: thrombocytocrit; PDW: platelet distribution width; GLU: glucose; FIB: fibrinogen; EOSP: percentage of eosinophils; GLB: globulin; PAB: prealbumin; MPV: mean platelet volume; TT: thrombin time; UA: uric acid; BUN: blood urea nitrogen; P: phosphorus; LY: lymphocyte count; ALP: alkaline phosphatase; K: kalium; CK: creatine kinase; CREA: creatinine; AST: glutamic oxalacetic transaminase; ALT: glutamic-pyruvic transaminase; GGT: gamma-glutamyl transpeptidase; TG: triglyceride; CA: calcium; MONP: percentage of monocyte; APTT: activated partial thromboplastin time; INR: international normalized ratio; RBCSD: red blood cell distribution width-standard deviation; PTP: percentage of prothrombin time; MCV: mean corpuscular volume; MCHC: mean corpuscular hemoglobin concentration; DD: d-dimer; BASP: percentage of basophil; CL: chloridion; DBIL: direct bilirubin

Fig. 3

The area under the receiver operating characteristic curve (AUC) of the random forest (A), GBM (B), GLM (C), and deep learning (D) models based on all the variables in the training set, internal validation set and the external test set. The area under the precision-recall curve (AUCPR) of the random forest (E), GBM (F), GLM (G), and deep learning (H) models based on all the variables in the training set, internal validation set and the external test set

Fig. 4

The area under the receiver operating characteristic curve (AUC) of the random forest (A), GBM (B), GLM (C), and deep learning (D) models based on the nine selected variables in the training set, internal validation set and the external test set. The area under the precision-recall curve (AUCPR) of the random forest (E), GBM (F), GLM (G), and deep learning (H) models based on the nine selected variables in the training set, internal validation set and the external test set

Fig. 5

Calibration curve of GBM model. A: GBM model calibration based on all the variables in the training set, internal validation set and the external test set. B: GBM model calibration based on the nine selected variables in the training set, internal validation set and the external test set. C: GBM model calibration in random sample set 1 based on the nine selected variables in the training set, internal validation set and the external test set. D: GBM model calibration in random sample set 2 based on the nine selected variables in the training set, internal validation set and the external test set. E: GBM model calibration in random sample set 3 based on the nine selected variables in the training set, internal validation set and the external test set

Fig. 6

The public internet calculator for RD discrimination by nine features. The application web server of GBM model with nine features available at http://www.empowerstats.net/pmodel/?m=31141_GBM9 for the RD prediction in patients with HM (A). Users could predict RD by submitting nine features into the text boxes. An example of a 44-year-old male participant with PCT of 0.18, GLU of 8.01, BASP of 0.21, GLB of 46, MPV of 10.02, UA of 0.26, and APTT of 33.50, who was enrolled in the Xuhui Central hospital in 2022 is demonstrated on this webpage (Fig. 6B). An example of a 36-year-old female participant with PCT of 0.24, GLU of 5.35, BASP of 0.41, GLB of 26.96, MPV of 10.15, UA of 0.31, and APTT of 33.37, who was enrolled in the EENT hospital in 2023 is demonstrated on this webpage (Fig. 6C).