Skip to main content

Table 1 Summary of included studies in this literature review

From: Machine learning for prediction of asthma exacerbations among asthmatic patients: a systematic review and meta-analysis

Studies (author, year)

Study design

Participants

Sample size (outcome%)

Outcome

Prediction duration

ML algorithms (n)

Validation methods

Generalization testa

Features in the final model

Performance measure in the validation dataset

Quality (PROBAST)

Bias

Applicability

Lieu, 1999 [16]

Retrospective

Asthma

7141 (HP 0.8%, ED 6.6%)

HP or ED for asthma

Next year

CART (1)

Split-sample validation

Y (in separate dataset)

14

Internal: Spe 0.836, Sen 0.49, PPV 0.185

High

High

Schatz, 2004 [17]

Retrospective

Asthma

8789 (5.5%)

HP or ED for asthma

Next year

LR (1)

-

Y (geographic)

3

External: Spe 0.92, Sen 0.254, NPV 0.932

High

High

Schatz, 2006 [18]

Retrospective

Asthma

1079 (9.5%)

HP or ED

Next year

LR (1)

-

Y (temporal)

1

External: Spe 0.541, Sen 0.646, NPV 0.96, PPV 0.083, AUC 0.59

High

High

Xu, 2011 [19]

Retrospective

Mild-moderate asthma

417 (30%)

Severe AE

Four years

RF (1)

Cross validation

Y (in separate sample)

164

External: AUC 0.66, Spe 0.6, Sen 0.66, NPV 0.81, PPV 0.74

High

High

van Vliet, 2017 [20]

Prospective

Asthma

574 samples (13.4%) from 94 patients (48%)

Moderate-severe AE

0–14 days, 0–21 days

RF (2)

Bagging

-

7

32 samples: 0–14 days (Spe 0.75, Sen 0.88, CCR 0.82, AUC 0.90) 48 samples: 0–21 days (Spe 0.67, Sen 0.85, CCR 0.65)

High

Low

Luo, 2020 [15]

Retrospective

Asthma

315,308 (3.6%)

HP or ED for asthma

Next year

XGBoost (1)

Cross validation

Y (temporal)

142

External: Spe 0.9193, Sen 0.5369, CCR 0.9031, AUC 0.859

High

High

Luo, 2020 [14]

Retrospective

Asthma

782,762 (2.42%)

HP or ED for asthma

Next year

XGBoost (1)

-

Y (temporal)

221

External: Spe 0.9091, Sen 0.5190, CCR 0.9008, AUC 0.820

High

High

Tong, 2021 [21]

Retrospective

Asthma

68,244 (1.7%)

HP or ED for asthma

Within 1 year

XGBoost (1)

Unclear

Y (temporal)

71

External: Spe 0.9091, Sen 0.7018, CCR: 0.906, AUC 0.902

High

High

Zein, 2021 [22]

Retrospective

Asthma

60,302 (nonsevere AE 32.8%, ED 2.9%, HP 1.5%)

Nonsevere AE (oral steroids burst), severe AE (HP or ED for asthma)

3 years (mean)

LR (3), RF (3), LGBM (3)

-

Y (20% of dataset, temporal)

56

20% of dataset: LGBM: Nonsevere (AUC 0.71, Spe 0.67, Sen 0.64), ED (AUC 0.88, Spe 0.76, Sen 0.84), HP (AUC 0.85, Spe 0.73, Sen 0.86))

High

High

External: LGBM: nonsevere (AUC 0.65), ED (AUC 0.86), HP (AUC 0.87))

Noble, 2021 [23]

Retrospective

Asthma

58,619 (1.65%)

HP for asthma

Within one year

LR (1)

-

Y (in separate dataset)

14

External: AUC 0.71, Spe 0.933, Sen 0.285

High

High

de Hond, 2022 [24]

Retrospective

Stable, mild-moderate asthma

92,787 daily measurements (0.2%) from 165 patients (30%)

Severe AE

2 days

XGBoost (1), one class SVM (1), LR (1)

Cross validation

Y (in separate dataset)

Unclear

External: XGBoost (AUC 0.81, Spe 0.89, Sen 0.59), LR: (AUC 0.88, Spe 0.82, Sen 0.84), one class SVM (Spe 0.87, Sen 0.34)

High

High

  1. aGeneralization tests included two types of validation. Some studies applied external validation, such as temporal and geographic validation. Some studies split a single dataset into a training dataset and a test dataset and used the latter to assess the generalizability of prediction models. We provided an example in Additional file 4
  2. HP hospitalization, ED emergency department visit, AE asthma exacerbation, Spe specificity, Sen sensitivity, PPV positive predictive value, NPV negative predictive value, AUC area under the curve, CCR correct classification rate, SVM support vector machine, LGBM Light gradient boosting machine, CART classification and regression trees, LR logistic regression, RF random forest