Table 3 Summary of development and performance outcomes reported for each article by sepsis-related endpoint

From: A scoping review on pediatric sepsis prediction technologies in healthcare

 

Sample Size

Participant age range

Sex (% male)

Setting

Time

Prevalence (%)

Validation

Features and ranking

Approach(es)

AUROC (95% CI)

Sensitivity (95% CI)

Specificity (95% CI)

Sepsis

Ehwerhemuepha et al.69

537837 (visits)

Median: 4; IQR: 8 years

53.60

ED

5–7 seconds from triage to predict

0.20

50% training, 15% validation, 35% testing; 10-fold cross-validation

47 (ranked)

Stochastic gradient boosting

0.976 (0.972–0.981)a

0.4414 (..)

0.999 (..)

Georgette et al.73

141510 (encounters)

Median: 4.5; IQR: (1.8–9.7) years

51.70

ED

3.2 hours before IV vasoactive infusion

3.40

66% training, 33% validation

4

Empirically derived shock index

0.69–0.78 (0.65–0.80)

0.835 (0.817–0.854)

0.428 (0.424–0.433)

Gilholm et al.46

3473 (patients)

Median: 2.1; IQR: (0.9–5.6) years

55.00

EDb

After clinical suspicion

15.10

10-fold cross-validation

16 (ranked)

Logistic regression (score)

0.80 (0.78–0.82)

0.90 (0.87–0.92)

0.51 (0.49–0.53)

Lamping et al.59

289 (patients)

Medians: 28–46; IQR: (4–120) months

53.30–62.50

ICU

After clinical suspicion

19.38

3-fold cross-validation

8 (ranked)

Random forest

0.78 (0.70–0.87)

..

..

Li et al.68

608 (patients)

28.45 ± 21.47 months

62.70–64.80

ICU

During patient encounter

50.80

70% training, 30% validation

7 (ranked for KD)

Logistic regression (score)

0.878 (..)

0.94 (..)

0.75 (..)

Marassi et al.71

12749 (patients)

0 to 18 years

..

ICU

At onset

2.11

80% training, 20% validation

11

XGBoost

..

..

..

Mawji et al.54

1612 (patients)

Medians: 13.1–16.5; IQR: (21.8–32.9) months

51.50

ED

At triage

5.16

80% training, 20% validation; 10-fold cross-validation

9 (ranked)

Logistic regression

0.86 (..)

0.59–0.91 (0.54–0.94)

0.50–0.92 (0.45–0.94)

Mercurio et al.62

35074 (encounters)

97.7% between 0.11 and 18.0 years

53.00

ED

During patient encounter

0.54

80% training, 20% validation; Cross-validation

20 (ranked)

Random forest

0.81 (..)a

0.93 (..)

0.84 (..)

Classification and regression tree

0.77 (..)a

0.85 (..)

0.70 (..)

Nguyen et al.58

3014 (admissions)

Median: 1.13; IQR: (0.15–4.30) years

56.30

ICU, Inpatient

Within the first 24, 48 hours with clinician suspicion

4.40

10-fold cross-validation

107

Tree augmented naïve bayes

0.866 (0.826–0.906),

0.867 (0.828–0.906)

0.463 (0.376–0.550),

0.469 (0.377–0.550)

0.949 (0.940–0.956),

0.953 (0.941–0.957)

Sanchez-Pinto et al.43

218839 (encounters)

Medians: 2.6–3.7; IQR: (0.6–9.4) years

47.00–51.50

ED, ICU

During patient encounter

1.20 (Mortality), 0.60–0.80 (Early mortality)

10-fold cross-validation

13

Stacked regression models (score)

0.71–0.92 (0.70–0.92),

0.79–0.96 (0.77–0.96)a

0.23–0.81 (0.20–0.85),

0.31–0.90 (0.26–0.93)

0.53–0.99 (0.52–1.00),

0.52–0.99 (0.52–0.99)

Schlapbach et al.56

4403 (patients)

Median: 2.1; IQR: (0.62–6.68) years

55.50

ICU

Within 1 hour of admission

5.20 (30-day mortality)

..

6 (ranked)

Logistic regression (score)

0.810–0.817 (0.779–0.855)

..

..

Sepanski et al.51

35586 (encounters)

Mean: 5.7; SD: 5.4 years

..

EDb

Within 24–48 hours

0.20

Split-sample validation

12

Logistic regression

0.849–0.918 (0.782–0.985)

0.703–0.770 (0.599–0.866)

0.977–0.981 (0.976–0.982)

Solé‐Ribalta et al.65

210 (patients)

Median: 42.7; IQR: (5.7–138.6) months

52.40

ED, Inpatient

After clinical suspicion

69.52

k-fold cross-validation

11 (ranked)

Logistic regression (score)

0.886 (0.845–0.927)

0.98 (..)

0.767 (..)

Solé‐Ribalta et al.66

266 (patients)

Median: 37.2; IQR: (4.2–127.8) months

53.80

ED

After clinical suspicion

64.66

k-fold cross-validation

4 (ranked)

Logistic regression (score)

0.825 (0.772–0.878)

0.88–0.90 (..)

0.45–0.62 (..)

Spaeder et al.60

1521 (patients)

Medians: 1.4–3.2; IQR: (0.3–7.4) years

50.3–53.7

ICU

Within 24 hours

10.20 (of admissions)

Cross-validation

37 (ranked)

Random forest

0.762 (0.728–0.789)

0.102 (..)

0.16 (..)

9 (ranked)

Logistic regression

0.735 (0.710–0.771)

0.215 (..)

0.204 (..)

Stephen et al.61

1425 (patients)

79.8–84.2% between 1 and > 18 years

53.70

ICUb

Within 24 hours, Anytime

1.40 (of encounters)

84.6% training, 15.4% validation; 40-fold cross-validation

13 (ranked)

Logistic regression (score)

..a

0.49 (0.37–0.61),

0.72 (0.60–0.81)

0.98 (0.98–0.98),

0.96 (0.96–0.96)

Yang et al.64

65 (patients)

Between 8.8–9.3 ± 4.6–4.9 years

42.9–56.7

ICU

After clinical suspicion

53.80

10-fold cross-validation

4

Logistic regression (score)

0.826–0.917 (..)

..

..

Ying et al.72

667 (patients)

0 to 10 years

..

ICU

After clinical suspicion

74.50

67% training, 33% validation; 10-fold cross-validation

18

Gradient boosting machine

0.943 (0.900–0.978)a

0.941 (..)

0.787 (..)

Severe Sepsis

Ehwerhemuepha et al.69

537837 (visits)

Median: 4; IQR: (8) years

53.60

ED

5–7 seconds from triage to predict

0.05

50% training, 15% validation, 35% testing; 10-fold cross-validation

47 (ranked)

Stochastic gradient boosting

0.99 (0.985–0.995)a

0.845 (..)

0.99 (..)

Kamaleswaran et al.55

493 (patients)

6 to 18 years

..

ICU

2, 8 hours earlier than existing alert

4.06

5-fold cross-validation

3 (ranked), 6 (ranked)

Logistic regression

0.77 (0.63–0.91),

0.56 (0.39–0.76)

0.55 (..),

0.393 (..)

0.874 (..),

0.771 (..)

14 (ranked)

Random forest

..

0.80 (..),

0.611 (..)

0.796 (..),

0.823 (..)

30

Convolutional neural network

..

0.75 (..),

0.76 (..)

0.83 (..),

0.81 (..)

Le et al.52

9486 (patients)

Median: 10; IQR: (5–14) years

50.39

Inpatient

4 hours before onset, at onset

1.06

4-fold cross-validation

7 (ranked), 6 (ranked)

Boosted ensembles of decision trees

0.718 (..),

0.916 (..)

0.75 (..),

0.75 (..)

0.70 (..),

0.94 (..)

Mercurio et al.62

35074 (encounters)

97.7% between 0.11 and 18.0 years

53.00

ED

During patient encounter

0.54

80% training, 20% validation; Cross-validation

20 (ranked)

Random forest

0.81 (..)a

0.93 (..)

0.84 (..)

Classification and regression tree

0.77 (..)a

0.85 (..)

0.70 (..)

Sepanski et al.51

35586 (encounters)

Mean: 5.7; SD: 5.4 years

..

EDb

Within 24–48 hours

0.20

Split-sample validation

12

Logistic regression

0.849–0.918 (0.782–0.985)

0.703–0.770 (0.599–0.866)

0.977–0.981 (0.976–0.982)

Septic Shock

Aviles-Robles et al.63

404 (febrile neutropenia episodes)

Median: 7.7; IQR (4.4–11.7) years

48.80

ED

During patient encounter

16.1 (of febrile neutropenia episodes)

80% training, 20% validation

9 (ranked)

Logistic regression

0.66 (0.56–0.76)

0.96 (..)

0.33 (..)

Georgette et al.73

141510 (encounters)

Median: 4.5; IQR: (1.8–9.7) years

51.70

ED

3.2 hours before IV vasoactive infusion

0.36

66% training, 33% validation

4

Empirically derived shock index

0.82–0.95 (0.65–0.97)

0.813 (0.752–0.874)

0.835 (0.832–0.838)

Liu et al.70

6161 (patients)

Mean: 6.04; SD: 5.75 years

55.70

ICU

Median early warning: 8.9 hours

17.58

70% training, 30% validation

26 (top 10 ranked)

XGBoost

0.90 (..)

0.84 (..)

0.82 (..)

Median early warning: 12.0 hours

26 (top 10 ranked)

Generalized linear model

0.87 (..)

0.83 (..)

0.75 (..)

Mercurio et al.62

35074 (encounters)

97.7% between 0.11 and 18.0 years

53.00

ED

During patient encounter

0.54

80% training, 20% validation; Cross-validation

20 (ranked)

Random forest

0.81 (..)a

0.93 (..)

0.84 (..)

Classification and regression tree

0.77 (..)a

0.85 (..)

0.70 (..)

Sanchez-Pinto et al.43

218839 (encounters)

Medians: 2.6–3.7; IQR: (0.6–9.4) years

47.00–51.50

ED, ICU

During patient encounter

1.20 (Mortality), 0.60–0.80 (Early mortality)

10-fold cross-validation

13

Stacked regression models (score)

0.71–0.92 (0.70–0.92),

0.79–0.96 (0.77–0.96)a

0.23–0.81 (0.20–0.85),

0.31–0.90 (0.26–0.93)

0.53–0.99 (0.52–1.00),

0.52–0.99 (0.52–0.99)

Schlapbach et al.56

4403 (patients)

Median: 2.1; IQR: (0.62–6.68) years

55.50

ICU

Within 1 hour of admission

5.20 (30-day mortality)

..

6 (ranked)

Logistic regression (score)

0.810–0.817 (0.779–0.855)

..

..

Scott et al.44

2464 (visits)

Medians: 5.6–6.0; IQR: (2.1–13.4) years

49.00–55.00

ED

At hospital arrival: temporal, geographic

11.40

10-fold cross validation, temporal and geographic validation

9 (ranked)

Elastic net regularization

0.75 (0.69–0.81), 0.87 (0.73–1.00)

0.82 (0.72–0.90),

0.90 (0.55–1.00)

0.48 (0.44–0.52),

0.32 (0.21–0.46)

Scott et al.45

2318 (visits)

Medians: 4.9–5.9; IQR: (10) years

52.00–55.00

ED, ICU

Within 2 hours of hospital arrival: temporal, geographic

8.50

Temporal and geographic validation

20 (ranked)

Logistic regression

0.83 (0.78–0.89),

0.83 (0.60–1.00)

0.84 (0.71–0.92),

0.80 (0.28–0.99)

0.65 (0.61–0.69),

0.40 (0.27–0.54)

Xiang et al.57

1238 (patients)

Median: 58.3; IQR: (25.8–111.0) months

59.60

Inpatient

4, 8, 12, 24 hours before onset

2.84 (of observation periods)

78% training, 22% validation

23 (ranked)

XGBoost

0.925 (..),

0.925 (..),

0.900 (..),

0.930 (..)

..

..

Ying et al.72

667 (patients)

0 to 10 years

..

ICU

After clinical suspicion

74.50

67% training, 33% validation; 10-fold cross-validation

18

Gradient boosting machine

0.943 (0.900–0.978)a

0.941 (..)

0.787 (..)

  1. (..) not available or reported, AUROC area under the receiver operating curve, CI confidence interval, ED emergency department, ICU intensive care unit, IQR interquartile range, KD Kawasaki Disease, SD standard deviation, XGBoost extreme gradient boosting.
  2. aAlso reports the area under the precision-recall curve (see Supplementary Data 1).
  3. bImplemented. Note that some articles appear more than once if they include multiple sepsis-related endpoints. See Supplementary Data 1 for Martinez et al.74 and Stephen et al.67.
Back to article page