Clinical prognostic scores for dengue fever: A systematic review

Table 2 Characteristics of included studies

Ref.	Period of data collection	Location	Study type	Model	Validation	Sample size
Chi et al[38], 2023	1 July 2015 to 30 November 2015	Tainan, Taiwan	Retrospective cross-sectional	Multivariate binary logistic regression with significant coefficient transformed to scores by inverse odds ratio	External validation (separate region and time)	701
Yang et al[37], 2023	August 15, 2019 and September 30, 2019	Dhaka, Bangladesh	Cross-sectional	CART model used on univariate and multivariate logistic regression models	Internal validation via split-sample with random assignment (80% training sample, 20% hold-out sample)	1090
Gayathri et al[10], 2023	Model: October 2019; Validation: September 2019 to January 2021	Chennai, India	Prospective cohort	Binary logistic regression to develop prediction severity model with forward stepwise method in 3 steps to identify 3 significant variables and Nagelkerke square to quantify influence of variables	Temporal validation on 2021 data (n = 312)	312
McBride et al[11], 2022	June 2019 to June 2021	Ho Chi Minh city, Vietnam	Prospective observational cohort	mSOFA score and delta excluding bilirubin calculated from day 0 and 2. Brier score rescaled from 0 to 1	Internal validation via bootstrap procedure with 500 resamples with replacement	124
Bhaskar et al[12], 2022	January 2016 to December 2020	Manipal, India	Prospective case cohort	Logistic regression model of significant variables	No validation	303
Srisuphanunt et al[23], 2022	2017 to 2019	Bangkok, Thailand	Retrospective cohort	Potential predictor tested for trend with nonparametric method	Internal validation (method not mentioned)	302
Sachdev et al[13], 2021	July 1, 2016 to December 31, 2019	New Deli, India	Prospective cohort	Multivariate logistic regression model to identify independent risk factors, stepwise entry of new terms into model	No validation	78
Marois et al[24], 2021	January 1, 2017 to July 31, 2017	New Caledonia	Retrospective cohort	Predictive model built using multiple logistic regression and descending stepwise analysis	Internal validation via k-fold cross-validation (k = 10)	383
Devarbhavi et al[25], 2020	January 2014 to December 2017	Bangalore, India	Retrospective cohort	MELD score, arterial pH, lactate used to generate ROC with C-statistics	No validation	36
Tangnararatchakit et al[26], 2020	2004 to 2018	Bangkok, Thailand	Retrospective cohort	Daily Dengue severity score created in Phase I (n = 191)	Temporal validation on Phase II (n = 51)	242
Lee et al[27], 2018	Kaohsiung Chang Gung Memorial Hospital: 2022 to 2015; Kaohsiung Medical University Hospital[2]: 2009 to 2013	Kaohsiung, Taiwan	Retrospective cohort	Multivariate logistic regression model and assigning points by dividing its regression coefficient by smallest coefficient in model (rounded to nearest whole number)	No validation	1068
Phakhounthong et al[28], 2018	October 12, 2009 to October 12, 2010	Siem Reap, Cambodia	Retrospective cohort	CART tree constructed with J48 algorithm to generate decision tree	Internal validation via 10-fold cross-validation by Weka sed to estimate out-of-sample accuracy (split data into 10, 9 for training, 1 for testing). Multiple rounds of cross-validation performed using different partitions	198
Park et al[14], 2018	Queen Sirikit National Institute of Child Health[3]: 1994 to 1997, 1999 to 2002, 2004 to 2007; Kamphaeng Phet Provincial Hospital[4]: 1994 to 1997	Bangkok, Thailand; Nai Mueang, Thailand	Prospective cohort	SEM using data from n = 257 with complete data	Internal validation via multiple imputation via Markov-chain Monte Carlo method to create 50 imputed datasets without missing data on n = 1244 to assess Sn	744
Md-Sani et al[29], 2018	September 8, 2022 to November 18, 2022	Kuala Lumpur, Malaysia	Retrospective cohort	Variable selection via 5-fold cross-validated Lasso regression used to build logistic regression model	Internal validation via cross-validation	199
Suwarto et al[30], 2018	January 2011 to March 2016	Jarkarta, Indonesia	Retrospective cohort	Dengue Score (Suwarto et al[18], 2016)	External validation	207
Hsieh et al[15], 2017	July 1, 2015 to December 31, 2015	Tainan, Taiwan	Prospective cohort	Univariate and multivariate with binary variables Cox model to identify predictive factors for mortality with cut-off values selected using Youden index	No validation	625
Huang et al[35], 2017	September 1, 2015 to December 31, 2015	Tainan, Taiwan	Case control	Univariate analysis and Multivariate logistic regression analysis to investigate independent predictors for 30-day mortality. Novel prediction score developed by assigning a score of 1 to each independent variable	Internal validation via bootstrapping method by generating 1000 hypothetical study population using random sampling from study sample	2358
Fernández et al[31], 2017	2009 to 2010	Tegucigalpa and San Pedro Sula, Honduras	Retrospective cohort	Univariable analysis and multivariable logistic regression analysis using forward stepwise selection to construct a predictive model for severe dengue	Internal validation via bootstrap technique (sampling with replacement using 320 individuals sampling 1000 times)	320
Nguyen et al[16], 2017	October 1, 2010 to December 31, 2013	Southern Vietnam	Prospective cohort	Logistic regression to develop prognostic model	Internal validation via "leave-one-site-out cross validation" (develop algorithm on all but 1 study site and validate using that study site) and Temporal validation	2060
Djossou et al[17], 2016	March 17, 2013 to September 30, 2013	Cayenne, French Guiana	Prospective cohort	Final model include variables with significant association in single covariable analysis	Internal validation via bootstrapping 1000 replications	806
Lee et al[32], 2016	Kaohsiung Chang Gung Memorial Hospital: July 1, 2002 to May 31, 2015; Kaohsiung Medical University Hospital[6]: 2009 to 2011	Kaohsiung, Taiwan	Retrospective cohort	Significant variables in univariate analysis entered into multivariate logistic regression and point assignment calculated by dividing regression coefficient by smallest coefficient in model	Temporal validation (model set before 31 Jul 2014 n = 1063, validation set after Aug 1 2014 n = 190)	1253
Suwarto et al[18], 2016	March 2010 to August 2015	Jarkarta, Indonesia	Prospective cohort	Variables entered into multiple regression analysis using backward selection algorithm to estimate coefficient and independent diagnostic predictors and converted into simplified risk score system	Validation published separately[30]	172
Lam et al[19], 2015	2003 to 2009	Ho Chi Minh City, Vietnam	Prospective cohort	Univariate and multivariate analysis via logistic regression and model simplified using stepwise backwards model selection based on Akaike Information Criterion	Temporal validation (model from n = 939 enrolled before 2009 and validated on 268 enrolled during 2009) and internal validation via repeated 10-fold cross-validation	1207
Pang et al[36], 2014	January 1, 2004 to December 31, 2008	Singapore	Case control	Univariate and multivariate conditional logistic regression performed to assess association	No validation	135
Pongpan et al[33], 2014	2007 to 2010	Phrae, Thailand; Lamphun, Thailand; Chiang Mai, Thailand	Retrospective cohort	Scoring system (Pongpan et al[34], 2013)	External validation	400
Pongpan et al[34], 2013	2007 to 2010	Nakorn Sawan, Thailand; Kampaeng Phet, Thailand; Uttaradit, Thailand	Retrospective cohort	Scoring system analysed by multivariable ordinal logistic regression and assigned item scores derived from coefficient transformation	Validation published separately[33]	777
Leo et al[20], 2013	January 2010 to September 2012	Singapore	Prospective cohort	Variables selected from World Health Organization[7] Warning Signs	External validation	499
Diaz-Quijano et al[21], 2010	Not reported	Bucaramanga, Colombia	Prospective cohort	Risk score based on independent predictors and risk group formed	No validation	729
Potts et al[22], 2010	Queen Sirikit National Institute of Child Health: 1994 to 1997, 1999 to 2002, 2004 to 2007; Kamphaeng Phet Provincial Hospital: 1994 to 1997	Bangkok, Thailand; Kamphaeng Phet, Thailand	Prospective cohort	CART analysis with age, gender, and clinical laboratory data to establish a diagnostic decision tree	Internal validation via k-fold cross validation method (k = 5) of each tree	582

CART: Classification and Regression Tree; MELD: Model for end-stage liver disease; mSOFA: Modified sequential organ failure assessment; ROC: Receiver operating characteristic.