Table 4 Properties of the attendances assigned to each of the explanation clusters (Fig. 4).

From: Using explainable machine learning to identify patients at risk of reattendance at discharge from emergency departments

Cluster

Count

Age (SD)

Reattendance rate (95% CI)

Predicted reattendance rate (95% CI)

Condition count

30-day visit count

Most important variables (mean absolute SHAP values)

1st

2nd

3rd

0

2723

55.4 (22.6)

5.9 (5.1–6.9)

6.8 (5.0–8.7)

2.9

0.0

30-day visit count

Triage complaint

Condition count

1

446

52.8 (21.4)

8.5 (6.3–11.5)

10.6 (7.1–14.0)

3.5

1.0

30-day visit count

Medical history

Triage complaint

2

1363

42.5 (19.2)

2.6 (1.9–3.6)

4.6 (4.4–5.0)

0.0

0.0

Condition count

30-day visit count

Triage complaint

3

671

40.9 (18.5)

4.0 (2.8–5.79)

5.6 (5.0–6.3)

0.0

0.0

Condition count

Diagnosis

30-day visit count

4

2761

41.5 (16.9)

1.8 (1.3–2.3)

3.8 (3.6–4.0)

0.0

0.0

Condition count

30 day visit count

Triage complaint

5

221

40.5 (15.9)

0.0 (0.0–1.7)

3.6 (3.4–3.7)

0.0

0.0

Condition count

30-day visit count

Triage complaint

6

321

39.7 (18.1)

5.6 (3.6–0.9)

5.9 (4.7–7.0)

0.0

1.0

Condition count

30-day visit count

Triage complaint

7

341

42.3 (17.4)

31.1 (26.4–36.2)

27.2 (13.6–40.7)

3.3

3.75

30-day visit count

Medical history

Condition count

  1. The count column displays the number of attendances in a given cluster. The age (in years), reattendance rate, predicted reattendance rate, condition count, and 30-day visit count column display the mean of the respective variable for all attendances in the cluster. Values in brackets in the age column display the standard deviation patient age in the respective cluster. The final three columns display the most important variables in making a decision, averaged across all attendances in a given cluster.