Table 3 Descriptive statistics and total and net information (scaled) of the 29 core predictors in the final model.

From: Development and validation of a machine learning model using electronic health records to predict trauma- and stressor-related psychiatric disorders after hospitalization with sepsis

Predictor

Domain

Development Sample, Median [IQR] or N (%)

Test Sample, Median [IQR] or N (%)

Total Information

Net Information

Pre-existing trauma- or stressor-related disorder (diagnosis present)

mental health diagnosis

3307 (1.5%)

1963 (1.6%)

1

1

Age (years)

demographics

71 [25]

70 [25]

0.064

0.028

Psychotherapeutic medication class (number of administrations)

medication

0 [3]

0 [3]

0.041

0.071

Comorbidity score

other admission data

43 [80]

48 [84]

0.032

0.048

Pre-existing mood disorder (diagnosis present)

mental health diagnosis

30,917 (14.2%)

16,421 (13.2%)

0.028

0.010

Oxygen saturation (mean value)

vital sign

96.94 [2.10]

96.85 [2.11]

0.024

0.018

White blood cell count (first value)

lab value

10.8 [7.20]

11 [7.30]

0.024

0.032

Systolic blood pressure (first value)

lab value

133 [35]

133 [35]

0.023

0.059

White blood cell count (last value)

lab value

8.30 [4.60]

8.30 [4.60]

0.023

0.033

Benzodiazepine (number of administrations)

medication

0 [0]

0 [0]

0.022

0.028

Heart rate (mean value)

vital sign

82.78 [18.14]

83.15 [18.27]

0.020

0.019

Pre-existing anxiety disorder (diagnosis present)

mental health diagnosis

18,218 (8.4%)

10,068 (8.1%)

0.020

0.025

Analgesic (number of administrations)

medication

4 [11]

3 [10]

0.019

0.014

Creatinine (first value)

lab value

0.97 [0.68]

1.00 [0.69]

0.017

0.022

Lactate (mean value)

lab value

1.40 [0.77]

1.50 [0.8]

0.017

0.017

Alert and oriented X3

mental status

179,393 (87.2%)

98,105 (87.2%)

0.015

0.046

Bicarbonate (first value)

lab value

25 [5]

26 [5]

0.015

0.019

Diastolic blood pressure (mean value)

vital sign

67.41 [12.44]

67.46 [12.96]

0.015

0.010

Length of stay (days)

hospitalization

3.10 [3.60]

3.00 [3.30]

0.014

0.030

Temperature (first value)

vital sign

98.40 [1.40]

98.40 [1.40]

0.014

0.043

Analgesic, opioid agonist (number of administrations)

medication

0 [3]

0 [2]

0.013

0.016

Analgesic, opioid (number of administrations)

medication

0 [4]

0 [3]

0.013

0.034

Arterial PaO2 (first value)

lab value

80 [48]

81 [47]

0.013

0.030

Hematocrit (maximum value)

lab value

37.50 [7.80]

37.70 [8.30]

0.013

0.017

Lactate (minimum value)

lab value

1.10 [0.60]

1.20 [0.70]

0.013

0.033

Height (inches)

other admission data

66.00 [6.01]

66.00 [6.01]

0.012

0.018

Analgesic, nonsteroidal anti-inflammatory (number of administrations)

medication

0 [0]

0 [0]

0.011

0.041

Sex (female)

demographics

117,768 (54.2%)

66,180 (53.1%)

0.011

0.023

Olanzapine (number of administrations)

medication

3 [6]

4 [5]

0.010

0.027

  1. Note. Total information denotes the relative contribution of predictors through direct associations with the outcome, and net information denotes the relative unique information contributed. Core predictors were identified using gradient boosting machines (GBM) 20% of the development sample (n = 29,822 hospitalizations). The remaining 80% of the development sample (n = 118,559 hospitalizations) was used to develop a GBM with only these 29 core predictors (selected as the optimal model).
Back to article page