Table 2 Summary of published studies using complex approaches to type 2 diabetes classification.

From: Precision subclassification of type 2 diabetes: a systematic review

Approach

Total n

% Female

Mean age/age range

% Race/ethnicity breakdown

% Prevalent or new onset T2D, n (%)

% Prospective or Cross sectional, n (%)

Machine learning approach

T2D subclassified Groups identified

Outcome category, n (%)

Overall quality, n (%)

References

Complex: Ahlqvist and directly replicated Ahlqvist clusters

(22 studies)

n = 88,197

43% female

55.3 years

81% Non-Hispanic White, 11% East Asian, 4% Hispanic, 3% South Asian, <1% Black, <1% Other

Prevalent (n = 11, 50%), New onset (n = 11, 50%)

Longitudinal (n = 8, 36.6%), Cross-sectional (n = 14, 63.6%)

100% k-means

SAID, SIDD, SIRD, MOD, MARD

Microvascular & macrovascular events (n = 9, 41%), Clinical and biochemical traits (n = 4, 18%)

Microvascular events only (n = 3, 13%), Glycaemia (n = 2, 9%), Macrovascular events only (n = 1, 5%), omic (n = 1, 5%), Other (n = 2, 9%)

Very Low (n = 1, 5%), Low (n = 3, 13%), Moderate (n = 18, 82%)

8,40,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108

Complex: Similar to Ahlqvist clusters

(13 studies)

n = 214,093

45% female

58.6 years

72% Non-Hispanic White, 12% Hispanic, 10% South Asian, 4% East Asian, 2% Black, <1% Native American, <1% Other

Prevalent (n = 11, 85%), New onset (n = 2, 15%) Longitudinal (n = 7, 54%), Cross sectional (n = 6, 46%)

a) Addition of complementary clinical variables (i.e., HDL, TG, waist circumference, uric acid, etc.)

b) Incorporating new clustering, i.e., self-normalising neural networks trained on k-means clustering.

c) Addition of ethnic-specific thresholds for BMI.

SAID, SIDD, SIRD, MOD, and MARD.

Five clusters: Older onset, Severe hyperglycaemia, Severe obesity, Younger Onset, and Insulin use.

Four clusters: 42% (older onset), 14% (poor glucose control), 24% (severe obesity), and 20% (younger-onset).

New subgroups MD, EOIDD, EOIRD, LOIDD, LOIRD.

Microvascular & macrovascular (n = 5, 38%), Microvascular events only (n = 3, 23%), Macrovascular events (n = 1, 8%), Glycaemia (n = 1, 8%), Other (n = 3, 23%)

Very Low (n = 1, 8%),

Low (n = 5, 38%)

Moderate (n = 7, 54%)

41,52,54,59,109,110,111,112,113,114,115,116,117

Complex: Simple clinical features

(4 studies)

n = 22,296

46.5% female

56 years

34% Non-Hispanic White, 25% Asian, 25% Middle Eastern, 9% Black, 5% Hispanic, 1% Native American/American Indian/Alaskan Native, <1% Other

Prevalent T2D (n = 3, 75%), New onset (n = 1, 25%)

Longitudinal (n = 3, 75%), Cross-sectional (n = 1, 25%)

50% k-means (n = 2), 50% other (n = 2)

variable

Mortality (n = 2, 50%), Cardiovascular events (n = 1, 25%), Clinical and biochemical traits (n = 1, 25%)

Moderate (n = 3, 75%), Low (n = 1, 25%)

118,119,120,121

Complex: Complex clinical features

(11 studies)

n = 386,889

46.8% female

60 years

57% Non-Hispanic White, 24% Asian, 8% Black, 6% Hispanic, <5% Other

Prevalent T2D (n = 9, 82%), New onset (n = 2, 18%)

Longitudinal (n = 7, 64%), Cross-sectional (n = 4, 36%)

variable

variable

Cardiovascular events (n = 4, 36%), Glycaemic control (n = 3, 27%), Complications other than CVD event (n = 2, 18%), Mortality (n = 1, 9%), Other (n = 1, 9%)

Moderate (n = 9, 82%), Low (n = 2, 18%)

29,30,122,123,124,125,126,127,128,129,130

Complex: Cardiovascular features

(2 studies)

n = 974

55.4% female

63 years

100% Non-Hispanic White

Prevalent T2D (n = 2, 100%)

Longitudinal (n = 2, 100%)

Factor analysis (clustering) (n = 1, 50%), Hierarchical clustering (n = 1, 50%)

3 and 4 clusters

Cardiovascular death and events (n = 2, 100%)

Moderate (n = 1, 50%), Low (n = 1, 50%)

31,32

Complex: Behavioural features

(2 studies)

n = 653

40.5% female 63.5 years

48% Non-Hispanic White, 50% Asian, 2% others

Prevalent (n = 1, 50%), New onset (n = 1, 50%)

Longitudinal (n = 1, 50%), Cross sectional (n = 1, 50%)

clustering, hierarchical clustering

2 and 4 clusters

Glycaemic control (n = 2, 100%)

Moderate (n = 1, 50%), Low (n = 1, 50%)

131,132

Complex: Glycemic features

(4 studies)

n = 67,064

42.8% female 62.5 years

40.6% Non-Hispanic White, 25% Asian, 25% Middle Eastern, 9.4% Other

Prevalent (n = 3, 75%), New onset (n = 1, 25%)

Longitudinal (n = 3, 75%), Cross sectional (n = 1, 25%)

50% k-means, 25% latent-class analysis, 25% hierarchical clustering

3 and 4 clusters

Glycaemic control (n = 2, 50%), Cardiovascular events (n = 2, 50%)

Moderate (n = 3, 75%), Very low (n = 1, 25%)

33,34,35,133

Complex: Genetics

(3 studies)

n = 42,952

100% Non-Hispanic White

Prevalent T2D (n = 3, 100%)

Cross sectional (n = 3, 100%)

Bayesian Non-negative Matrix Factorisation (n = 2, 67%), Hierarchical clustering (n = 1, 33%)

5 clusters of variant-trait associations; 3 clusters of skeletal dysregulated genes/pathways in people with diabetes

Coronary artery disease, stroke, renal disease

Moderate (n = 2, 67%), Low (n = 1, 33%)

10,37,38

Complex: Hormonal (1 study)

n = 96 participants

53% female

62 years

100% Non-Hispanic White

New onset (n = 1, 100%)

Cross sectional (n = 1, 100%)

Two-step cluster analysis

using log-likelihood distance measures

Two clusters (cluster 1: low GLP-1 and Ghrelin; cluster 2: high GLP-1 and Ghrelin)

Glycemia (n = 1, 100%)

Moderate (n = 1, 100%)

36

  1. T2D type 2 diabetes, GAD glutamic decarboxylase antibody, UCPCR urine C-peptide to creatinine ratio, BMI body mass index, CVD cardiovascular disease, OGTT oral glucose tolerance test, CKD chronic kidney disease, LDL low density lipoprotein cholesterol, SAID severe autoimmune diabetes, SIDD severe insulin deficient diabetes, SIRD severe insulin resistant diabetes, MOD mild obesity-related diabetes, MARD mild age-related diabetes, MD mild diabetes, EOIDD early-onset insulin deficient diabetes, EOIRD early-onset insulin resistant diabetes, LOIDD late-onset insulin deficient diabetes, LOIRD late-onset insulin resistant diabetes.