The predictive effect of different machine learning algorithms for pressure injuries in hospitalized patients: A network meta-analyses
Chaoran Qu,Weixiang Luo,Zhixiong Zeng,Xiaoxu Lin,Xuemei Gong,Xiujuan Wang,Yu Zhang,Yun Li
DOI: https://doi.org/10.1016/j.heliyon.2022.e11361
IF: 3.776
2022-11-03
Heliyon
Abstract:Background Pressure injury has always been a focus and difficulty of nursing. With the development of nursing informatization, a large amount of structured and unstructured data has been generated, and it is difficult for traditional methods to utilize these data. With the intersection of artificial intelligence and nursing, it has become a new trend to apply machine learning algorithms to build pressure injury prediction models to manage pressure injuries. However, there is no evidence on the effectiveness of the method and which of a large number of algorithms for machine learning is more applicable to pressure injuries. Objective This review aims to systematically synthesize existing evidence to determine the effectiveness of applying machine learning algorithms for pressure injury management, to further evaluate and compare pressure injury prediction models constructed by numerous machine learning algorithms, and to derive evidence for the best algorithms for predicting and managing pressure injuries. Design Systematic review and network meta-analysis. Methods A systematic electronic search was conducted in the EBSCO, Embase, PubMed, and Web of Science databases. We included all retrospective diagnostic accuracy trials and prospective diagnostic accuracy trials constructing a predictive model by machine learning for pressure injuries up to December 2021. Two review authors independently selected relevant studies and extracted data using the Cochrane handbook for systematic reviews of diagnostic test accuracy. The network meta-analysis was conducted using statistical software R and STATA. The certainty of the evidence was rated using the QUADAS-2 tool. Result Twenty-five clinical diagnostic trials with a total of 237397 participants were identified in this review. The results of our study revealed that pressure injury machine learning models can effectively predict these injuries. Combining the algorithms separately yields the main results: decision trees (sensitivity: 0.66, 95% CI: 0.42 to 0.84, specificity: 0.90, 95% CI: 0.78 to 0.96, diagnostic odds ratio [DOR]: 18, 95% CI: 7 to 49, AUC: 0.88, 95% CI: 0.85 to 0.91), logistic regression (sensitivity: 0.71, 95% CI: 0.60 to 0.80, specificity: 0.83, 95% CI: 0.75 to 0.89, DOR: 12, 95% CI: 9 to 17, AUC: 0.84, 95% CI: 0.81 to 0.87), neural networks (sensitivity: 0.73, 95% CI: 0.55 to 0.86, specificity: 0.78, 95% CI: 0.65 to 0.87, DOR: 9, 95% CI: 5 to 19, AUC: 0.82, 95% CI: 0.79 to 0.85), random forests (sensitivity: 0.72, 95% CI: 0.26 to 0.95, specificity: 0.96, 95% CI: 0.80 to 0.99, DOR: 56, 95% CI: 3 to 1258, AUC: 0.95, 95% CI: 0.93 to 0.97), support vector machines (sensitivity: 0.81, 95% CI: 0.69 to 0.90, specificity: 0.81, 95% CI: 0.59 to 0.93, DOR: 19, 95% CI: 6 to 54, AUC: 0.88, 95% CI: 0.85 to 0.90). According to the analysis of ROC and AUC values, random forest is the best algorithm for the prediction model of pressure injury. Conclusions This review revealed that machine learning algorithms are generally effective in predicting pressure injuries, and after data merging, the random forest algorithm is the best algorithm for pressure injury prediction. Further well-designed diagnostic controlled trials are recommended to strengthen the current evidence. Registration number (PROSPERO) : CRD42021276993.