Table 1 Coarse clustering results.

From: Novel architecture for gated recurrent unit autoencoder trained on time series from electronic health records enables detection of ICU patient subgroups

Coarse cluster name

Descriptive statistics

Prevalent diagnoses

Enriched diagnoses

Prevalent procedures

Enriched procedures

Population

N = 49599

43.9% female

Age: 66 years (53, 78)

ICU stay

- all: 2 days (1, 4)

- survivors: 2 days (1, 4)

- deceased: 3 days (1, 6)

28-day mortality: 13.6%

401.9 - Unspecified essential hypertension (41.3%),

414.0 Coronary atherosclerosis (27.9%),

427.3 Atrial fibrillation and flutter (26.6%),

428.0 - Congestive heart failure, unspecified (26.2%),

250.0 Diabetes mellitus without mention of complication or

manifestation classifiable to 250.1–250.9 (19.7%),

518.8 Other disease of lung (19.5%),

584.9 - Acute kidney failure, unspecified (18.2%),

272.4 - Other and unspecified hyperlipidemia (17.4%),

V58.6 Long-term (current) drug use (14.1%),

530.8 Other specified disorders of esophagus (13.2%),

599.0 - Urinary tract infection, site not specified (13.1%),

V45.8 Other postprocedural status (12.3%),

403.9 Unspecified hypertensive kidney disease (12.1%),

272.0 - Pure hypercholesterolemia (11.8%),

995.9 Systemic inflammatory response syndrome (SIRS) (10.8%)

–

38.93 - Venous catheterization, not elsewhere classified (25.1%),

96.04 - Insertion of endotracheal tube (17.0%),

96.71 - Continuous invasive mechanical ventilation for less than 96

consecutive hours (15.6%),

96.6 - Enteral infusion of concentrated nutritional substances

(14.9%),

99.04 - Transfusion of packed cells (13.8%),

39.61 - Extracorporeal circulation auxiliary to open heart surgery

(13.6%),

96.72 - Continuous invasive mechanical ventilation for 96 consecutive

hours or more (11.2%),

88.56 - Coronary arteriography using two catheters (10.2%),

36.15 - Single internal mammary-coronary artery bypass (8.8%),

38.91 - Arterial catheterization (8.7%),

88.72 - Diagnostic ultrasound of heart (6.6%),

37.22 - Left heart cardiac catheterization (6.4%),

39.95 - Hemodialysis (6.4%),

99.15 - Parenteral infusion of concentrated nutritional substances

(5.9%),

33.24 - Closed [endoscopic] biopsy of bronchus (5.3%)

–

Cluster 1

N = 19147

78.1% female

Age: 69 years (55, 81)

ICU stay

- all: 3 days (1, 6)

- survivors: 3 days (1, 6)

- deceased: 3 days (1, 7)

28-day mortality: 23.5%

\(\sim\) 401.9 - Unspecified essential hypertension (40.3%, -3.9%),

+ 518.8 Other disease of lung (29.2%, +118.1%),

+ 428.0 - Congestive heart failure, unspecified (28.5%, +15.1%),

\(\sim\) 427.3 Atrial fibrillation and flutter (27.2%, +3.9%),

+ 584.9 - Acute kidney failure, unspecified (21.1%, +28.4%),

− 414.0 Coronary atherosclerosis (20.3%, -38.0%),

\(\sim\) 250.0 Diabetes mellitus without mention of complication or

manifestation classifiable to 250.1-250.9 (19.8%, +1.3%),

+ 599.0 - Urinary tract infection, site not specified (18.2%,

+84.1%),

\(\sim\) 272.4 - Other and unspecified hyperlipidemia (17.0%, -3.7%),

+ V58.6 Long-term (current) drug use (16.2%, +26.4%),

+ 995.9 Systemic inflammatory response syndrome (SIRS) (14.9%,

+81.6%),

+ 244.9 - Unspecified acquired hypothyroidism (13.7%, +85.6%),

\(\sim\) 530.8 Other specified disorders of esophagus (13.4%, +2.8%)

+ 518.8 Other disease of lung (29.2%, +118.1%),

− 414.0 Coronary atherosclerosis (20.3%, -38.0%),

− 600.0 Hypertrophy (benign) of prostate (1.0%, -82.9%),

+ 599.0 - Urinary tract infection, site not specified (18.2%,

+84.1%),

+ 733.0 Osteoporosis (6.9%, +219.8%),

+ 785.5 Shock without mention of trauma (12.7%, +112.0%),

+ 995.9 Systemic inflammatory response syndrome (SIRS) (14.9%,

+81.6%),

+ 038.9 - Unspecified septicemia (10.9%, +106.3%),

+ 244.9 - Unspecified acquired hypothyroidism (13.7%, +85.6%),

+ 294 Persistent mental disorders due to conditions classified

elsewhere (6.6%, +174.6%),

+ 518.5 Pulmonary insufficiency following trauma and surgery (6.2%,

+131.0%),

+ 276.2 - Acidosis (11.9%, +67.5%),

+ 507.0 - Pneumonitis due to inhalation of food or vomitus (9.9%,

+72.4%)

+ 38.93 - Venous catheterization, not elsewhere classified (31.8%,

+52.1%),

+ 96.04 - Insertion of endotracheal tube (26.7%, +144.9%),

+ 96.6 - Enteral infusion of concentrated nutritional substances

(23.0%, +132.5%),

+ 96.71 - Continuous invasive mechanical ventilation for less than

96 consecutive hours (21.7%, +84.7%),

+ 96.72 - Continuous invasive mechanical ventilation for 96

consecutive hours or more (19.7%, +240.9%),

− 99.04 - Transfusion of packed cells (13.0%, -9.0%),

+ 38.91 - Arterial catheterization (12.7%, +104.2%),

+ 99.15 - Parenteral infusion of concentrated nutritional substances

(7.9%, +69.6%),

− 39.61 - Extracorporeal circulation auxiliary to open heart surgery

(7.6%, -56.0%),

+ 39.95 - Hemodialysis (7.3%, +26.8%),

− 88.56 - Coronary arteriography using two catheters (7.1%, -42.0%),

+ 33.24 - Closed [endoscopic] biopsy of bronchus (7.0%, +68.0%),

+ 31.1 - Temporary tracheostomy (6.1%, +236.4%)

+ 96.04 - Insertion of endotracheal tube (26.7%, +144.9%),

+ 96.6 - Enteral infusion of concentrated nutritional substances

(23.0%, +132.5%),

+ 96.72 - Continuous invasive mechanical ventilation for 96

consecutive hours or more (19.7%, +240.9%),

− 36.15 - Single internal mammary-coronary artery bypass (3.9%,

-67.4%),

− 39.61 - Extracorporeal circulation auxiliary to open heart surgery

(7.6%, -56.0%),

+ 96.71 - Continuous invasive mechanical ventilation for less than

96 consecutive hours (21.7%, +84.7%),

+ 38.93 - Venous catheterization, not elsewhere classified (31.8%,

+52.1%),

+ 31.1 - Temporary tracheostomy (6.1%, +236.4%),

+ 38.91 - Arterial catheterization (12.7%, +104.2%),

+ 43.11 - Percutaneous [endoscopic] gastrostomy [PEG] (5.1%,

+198.1%),

− 36.12 - (Aorto)coronary bypass of two coronary arteries (1.8%,

-64.2%),

− 88.56 - Coronary arteriography using two catheters (7.1%, -42.0%),

− 37.22 - Left heart cardiac catheterization (4.0%, -49.4%)

Cluster 2

N = 30452

22.3% female

Age: 64 years (52, 76)

ICU stay

- all: 2 days (1, 4)

- survivors: 2 days (1, 4)

- deceased: 3 days (1, 6)

28-day mortality: 7.3%

\(\sim\) 401.9 - Unspecified essential hypertension (41.9%, +4.0%),

+ 414.0 Coronary atherosclerosis (32.7%, +61.4%),

\(\sim\) 427.3 Atrial fibrillation and flutter (26.2%, -3.8%),

− 428.0 - Congestive heart failure, unspecified (24.7%, -13.1%),

\(\sim\) 250.0 Diabetes mellitus without mention of complication or

manifestation classifiable to 250.1-250.9 (19.6%, -1.3%),

\(\sim\) 272.4 - Other and unspecified hyperlipidemia (17.7%, +3.8%),

− 584.9 - Acute kidney failure, unspecified (16.4%, -22.1%),

+ 272.0 - Pure hypercholesterolemia (13.8%, +56.6%),

+ V45.8 Other postprocedural status (13.5%, +30.4%),

− 518.8 Other disease of lung (13.4%, -54.1%),

\(\sim\) 530.8 Other specified disorders of esophagus (13.0%, -2.7%),

− V58.6 Long-term (current) drug use (12.8%, -20.9%),

\(\sim\) 403.9 Unspecified hypertensive kidney disease (12.0%, -2.9%)

− 518.8 Other disease of lung (13.4%, -54.1%),

+ 414.0 Coronary atherosclerosis (32.7%, +61.4%),

+ 600.0 Hypertrophy (benign) of prostate (5.6%, +485.5%),

− 599.0 - Urinary tract infection, site not specified (9.9%, -45.7%),

− 733.0 Osteoporosis (2.2%, -68.7%),

− 785.5 Shock without mention of trauma (6.0%, -52.8%),

− 995.9 Systemic inflammatory response syndrome (SIRS) (8.2%,

-44.9%),

− 038.9 - Unspecified septicemia (5.3%, -51.5%),

− 244.9 - Unspecified acquired hypothyroidism (7.4%, -46.1%),

− 294 Persistent mental disorders due to conditions classified

elsewhere (2.4%, -63.6%),

− 518.5 Pulmonary insufficiency following trauma and surgery (2.7%,

-56.7%),

− 276.2 - Acidosis (7.1%, -40.3%),

− 507.0 - Pneumonitis due to inhalation of food or vomitus (5.7%,

-42.0%)

− 38.93 - Venous catheterization, not elsewhere classified (20.9%,

-34.3%),

+ 39.61 - Extracorporeal circulation auxiliary to open heart surgery

(17.3%, +127.3%),

+ 99.04 - Transfusion of packed cells (14.3%, +9.9%),

+ 88.56 - Coronary arteriography using two catheters (12.2%, +72.4%),

+ 36.15 - Single internal mammary-coronary artery bypass (11.9%,

+206.4%),

− 96.71 - Continuous invasive mechanical ventilation for less than

96 consecutive hours (11.7%, -45.9%),

− 96.04 - Insertion of endotracheal tube (10.9%, -59.2%),

− 96.6 - Enteral infusion of concentrated nutritional substances

(9.9%, -57.0%),

+ 37.22 - Left heart cardiac catheterization (7.9%, +97.6%),

+ 88.72 - Diagnostic ultrasound of heart (7.3%, +31.2%),

− 38.91 - Arterial catheterization (6.2%, -51.0%),

+ 37.23 - Combined right and left heart cardiac catheterization

(6.1%, +55.0%),

− 96.72 - Continuous invasive mechanical ventilation for 96

consecutive hours or more (5.8%, -70.7%)

− 96.04 - Insertion of endotracheal tube (10.9%, -59.2%),

− 96.6 - Enteral infusion of concentrated nutritional substances

(9.9%, -57.0%),

− 96.72 - Continuous invasive mechanical ventilation for 96

consecutive hours or more (5.8%, -70.7%),

+ 36.15 - Single internal mammary-coronary artery bypass (11.9%,

+206.4%),

+ 39.61 - Extracorporeal circulation auxiliary to open heart surgery

(17.3%, +127.3%),

− 96.71 - Continuous invasive mechanical ventilation for less than

96 consecutive hours (11.7%, -45.9%),

− 38.93 - Venous catheterization, not elsewhere classified (20.9%,

-34.3%),

− 31.1 - Temporary tracheostomy (1.8%, -70.3%),

− 38.91 - Arterial catheterization (6.2%, -51.0%),

− 43.11 - Percutaneous [endoscopic] gastrostomy [PEG] (1.7%, -66.4%),

+ 36.12 - (Aorto)coronary bypass of two coronary arteries (5.1%,

+179.5%),

+ 88.56 - Coronary arteriography using two catheters (12.2%, +72.4%),

+ 37.22 - Left heart cardiac catheterization (7.9%, +97.6%)

  1. We show basic descriptive statistics of the two clusters as well as the population. Frequent ICD-9 diagnoses or procedures are shown if they are enriched within a cluster (“+” for positively enriched, “−” for negatively enriched) and are sorted by the degree of enrichment within the cluster from high to low. “Prevalent” codes are sorted by the prevalence. Note that often, negatively enriched codes can be just as important in understanding a cluster as positively enriched codes. After each code, we show its support within the cluster (i.e. the fraction of admissions within the cluster that exhibit all diagnoses in the itemset) and the ratio between support in the cluster and support in the complement. Distributions of frequent ICD-9 codes vary between clusters. While patients in Cluster 1 suffer from acute conditions (lung diseases, heart failure, acute kidney failure, septicemia) more often, patients in Cluster 2 are more likely to exhibit chronic conditions. Additionally, differences between the clusters in terms of age and sex distribution, length of ICU stay, and mortality can be observed. Apart from age and sex, none of this data is supplied as input to our model. Thus, it follows these differences stem from information captured in the learned feature space.
Back to article page