Diabetologia (2022) 65:1770–1781

https://doi.org/10.1007/s00125-021-05625-x

REVIEW

A novel diabetes typology: towards precision diabetology

from pathogenesis to treatment
Christian Herder 1,2,3 & Michael Roden 1,2,3

Received: 25 June 2021 / Accepted: 4 October 2021 / Published online: 4 January 2022
# The Author(s) 2022

Abstract
The current classification of diabetes, based on hyperglycaemia, islet-directed antibodies and some insufficiently defined
clinical features, does not reflect differences in aetiological mechanisms and in the clinical course of people with
diabetes. This review discusses evidence from recent studies addressing the complexity of diabetes by proposing novel
subgroups (subtypes) of diabetes. The most widely replicated and validated approach identified, in addition to severe
autoimmune diabetes, four subgroups designated severe insulin-deficient diabetes, severe insulin-resistant diabetes, mild
obesity-related diabetes and mild age-related diabetes subgroups. These subgroups display distinct patterns of clinical
features, disease progression and onset of comorbidities and complications, with severe insulin-resistant diabetes show-
ing the highest risk for cardiovascular, kidney and fatty liver diseases. While it has been suggested that people in these
subgroups would benefit from stratified treatments, RCTs are required to assess the clinical utility of any reclassification
effort. Several methodological and practical issues also need further study: the statistical approach used to define
subgroups and derive recommendations for diabetes care; the stability of subgroups over time; the optimal dataset
(e.g. phenotypic vs genotypic) for reclassification; the transethnic generalisability of findings; and the applicability in
clinical routine care. Despite these open questions, the concept of a new classification of diabetes has already allowed
researchers to gain more insight into the colourful picture of diabetes and has stimulated progress in this field so that
precision diabetology may become reality in the future.

Keywords Clustering . Complications . Diabetes subgroups . Personalised medicine . Precision medicine . Reclassification .
Review

Abbreviations
ADOPT A Diabetes Outcome Progressive Trial
* Christian Herder AHEAD Action for Health in Diabetes
christian.herder@ddz.de ANDIS All New Diabetics in Scania
* Michael Roden ANGPTL8 Angiopoietin-like protein 8
michael.roden@ddz.de CAN Cardiovascular autonomic neuropathy
1
CASP-8 Caspase-8
Institute for Clinical Diabetology, German Diabetes Center CKD Chronic kidney disease
(Deutsches Diabetes-Zentrum/DDZ), Leibniz Center for Diabetes
Research at Heinrich-Heine-University Düsseldorf, DSPN Distal sensorimotor polyneuropathy
Düsseldorf, Germany EN-RAGE S100 calcium-binding protein A12
2
Department of Endocrinology and Diabetology, Medical Faculty and GDS German Diabetes Study
University Hospital Düsseldorf, Heinrich-Heine-University GLP-1RA Glucagon-like peptide-1 receptor agonist
Düsseldorf, Düsseldorf, Germany hsCRP High-sensitivity C-reactive protein
3
German Center for Diabetes Research (DZD), Partner Düsseldorf, MARD Mild age-related diabetes
München-Neuherberg, Germany MOD Mild obesity-related diabetes
Diabetologia (2022) 65:1770–1781 1771

NAFLD Non-alcoholic fatty liver disease precision medicine approach that is required for type 1 and
PRS Polygenic risk score type 2 diabetes, which are both polygenic and multifactorial
SAID Severe autoimmune diabetes [12].
SGLT2i Sodium–glucose cotransporter 2 inhibitor In the context of type 1 and type 2 diabetes, the ultimate
SIDD Severe insulin-deficient diabetes purpose of precision diabetology is the development of strat-
SIRD Severe insulin-resistant diabetes ified prevention and treatment for subgroups of people with
VNDS Verona Newly Diagnosed Type 2 Diabetes different risk profiles. These options range from refined
Study screening and monitoring intervals, recommendations for
tailored lifestyle interventions, to targeted but not
individualised drug treatment. Clinical benefits envisaged
include fewer adverse effects and ideally a delay of the onset
of diabetes and its complications, lower morbidity and mortal-
Rationale for diabetes reclassification ity and an economic use of resources [13].
The aim of this review is to provide an up-to-date, concise
The observation that people with diabetes have different overview of studies on diabetes reclassification, their implica-
phenotypes has led to repeated attempts to classify the main tions, and also inherent practical and methodological chal-
diabetes types [1, 2]. Electronic supplementary material lenges, with a specific focus on recent definitions of
(ESM) Table 1 summarises the key efforts in this evolution. subgroups of type 2 diabetes and the risk of complications in
People with diabetes show a broad variation in the main these subgroups. With respect to subgroups and endotypes of
features of diabetes (i.e. insulin resistance and beta cell type 1 diabetes, we would like to refer the reader to recent
dysfunction [3–6]) as a result of the combined effects of excellent reviews covering aspects of precision diabetology
(epi)genetic, environmental and lifestyle factors and their for this diabetes type [8, 14].
different contributions in different individuals. A recently
proposed ‘palette model’ conceptualises the interaction of
these factors [3]. People at risk of diabetes may have impair- Variability of disease presentation
ments in multiple processes such as islet development, islet and progression
function, autoimmunity, inflammation, insulin sensitivity,
incretin activity and adipose tissue function (considered as One approach to study the heterogeneity of diabetes relies on
‘base colours’). Every individual is positioned somewhere cohorts of people included at or shortly after the diagnosis of
within the spectrum of the phenotypic variation of each trait diabetes [5]. Even though the duration of hyperglycaemia
as determined by their genetic variation and non-genetic expo- before diabetes diagnosis is unknown, these cohorts allow
sures modifying these processes, and the sum (or mixture) of the investigation of clinical characteristics that are not yet
all trait variations represents the overall estimate of metabolic confounded by long-term excessive hyperglycaemia and phar-
health and diabetes status [3]. However, this pathophysiologi- macological treatment. Examples are the German Diabetes
cal heterogeneity is not captured by current position state- Study (GDS [5]), the Verona Newly Diagnosed Type 2
ments and guidelines for diagnosis and treatment of diabetes Diabetes Study (VNDS [15]) and the All New Diabetics in
[7, 8]. Scania (ANDIS [6]) cohorts.
Differences in risk factors and pathophysiological mecha- The comprehensive phenotyping in the GDS, using gold-
nisms are thought to drive the heterogeneity in preclinical standard methodology, demonstrated large interindividual
abnormalities, prevalence of comorbidities and clinical differences in people within 12 months of their diagnosis of
complications already seen at diagnosis of diabetes [5, 9]. diabetes regarding insulin sensitivity, beta cell function, islet-
People with diabetes further vary in the progression of their directed autoantibodies, blood lipids and BP [5, 16]. The vari-
disease and in the incidence of diabetes-related complications ability in disease severity or progression is reflected by differ-
despite comparable glycaemic control. ences between subgroups of type 2 diabetes regarding
Any reclassification effort should be seen as an example of diabetes-related complications such as chronic kidney disease
precision medicine or ‘precision diabetology’ aiming to (CKD), distal sensorimotor polyneuropathy (DSPN), cardio-
deconstruct the heterogeneity of diabetes. Advances in the vascular autonomic neuropathy (CAN), retinopathy and non-
management of monogenic forms of diabetes (neonatal diabe- alcoholic fatty liver disease (NAFLD) [5, 16, 17].
tes, MODY) represent a successful proof-of-concept for a The VNDS enrols people with type 2 diabetes within
reclassification of diabetes [10]. Currently, however, relative- 6 months of their diagnosis. Assessment of diabetes-related
ly few people with diabetes are affected by monogenic diabe- complications indicated a high variability in their presentation
tes so this may serve as an example for personalised medicine and progression [15, 18]; the prevalence of one or more
based on mutations in single genes [11] rather than the diabetes-related complications (CVD, nephropathy, DSPN,
1772 Diabetologia (2022) 65:1770–1781

CAN, retinopathy) already present at study enrolment was related to insulin resistance. Individuals with high genetic risk
found to be 49.2% [9]. scores in the respective clusters also differed in obesity, lipids,
The ANDIS cohort includes incident cases of diabetes and hypertension, kidney function and CVD [22].
reported a marked prevalence of NAFLD at baseline From a clinical perspective, clustering algorithms based on
(although only based on surrogate measurement by alanine available patient data would be highly attractive. One study
aminotransferase) and the development of CKD, retinopathy used latent-class trajectory analysis based on mixed-meal
and CVD in the first years after the diagnosis of diabetes [6]. tolerance tests in people with newly diagnosed type 2 diabetes
In addition, people with diabetes differ with respect to [25] and identified three subgroups based on their glucose
inherited factors. The application of a polygenic risk score response patterns. Thus, this method represents another clas-
(PRS) based on >136,000 variants in the UK Biobank demon- sification approach closely related to insulin resistance and
strated a prevalence of type 2 diabetes of 1.2% and 11.2% in insulin secretion as the pathophysiological hallmarks of type
the lowest and highest 2.5% of the PRS distribution, respec- 2 diabetes.
tively, revealing an almost tenfold difference [19].
Taken together, these studies highlight the substantial vari- Diabetes subgroups The most frequently replicated study in
ability in the pathogenic and clinical characteristics of the this field used both hierarchical and k-means clustering in
large population of people commonly designated as having Swedish people with newly diagnosed diabetes, with the
type 2 diabetes. following six variables as input: GAD antibodies; age at diag-
nosis; BMI at diagnosis; HbA1c; and HOMA-2 estimates of
insulin resistance and beta cell function calculated from
Novel subtypes of diabetes reflecting fasting glucose and C-peptide [6]. The resulting subgroups
differences in disease development (subtypes) were designated as severe autoimmune diabetes
and progression (SAID), severe insulin-deficient diabetes (SIDD), severe
insulin-resistant diabetes (SIRD), mild obesity-related diabe-
Reclassification methods From a methodological perspective, tes (MOD) and mild age-related diabetes (MARD) [6]
different clustering algorithms have been used to reclassify (Table 1). This concept has been replicated in cohorts from
people with diabetes [20–25]. Topology-based analysis [21] Europe, North America and Asia despite varying disease dura-
and Bayesian non-negative matrix factorisation clustering tion since diabetes diagnosis [26–32]. The SAID subgroup
[22] are widely applied procedures for discovering groups of comprises people who are otherwise classified as having type
related observations (e.g. subgroups of people with diabetes) 1 diabetes (including those previously termed latent autoim-
using high-dimensional data such as electronic medical mune diabetes of adults), whereas SIDD, SIRD, MOD and
records or omics data. Cluster analysis based on the k-means MARD represent novel entities of type 2 diabetes. The
or the partitioning around medoids methods [6, 23] break subgroups also differ with respect to prevalence and/or risk
large datasets up into subgroups by minimising the distance of complications (Table 1). In line with the predominant insu-
between data points labelled to be in a cluster and a point lin deficiency, ketoacidosis at diagnosis is most frequent in
designated as the centre of that cluster. These methods require SAID and SIDD [6]. Retinopathy, DSPN and CAN are
that the optimal number of clusters (k) needs to be known a observed most often in SIDD [6, 26], while CKD and
priori (i.e. evaluated with other methods). In contrast, latent- NAFLD are most prevalent in SIRD [6, 26, 29]. Adjusted risk
class trajectory analysis is a longitudinal analysis method ratios for prevalent erectile dysfunction are highest for SIDD
using repeated measures of dependent variables as a function and SIRD [33]. Although the subgroups differ in their cardio-
of time to identify subgroups of people who differ in trajecto- vascular risk, these differences did not remain statistically
ries (e.g. in glucose response curves) [24, 25]. significant after adjustment for age and sex in the ANDIS
In addition to these methods, reclassification studies made cohort [6] or after more comprehensive adjustment for multi-
use of datasets that differed widely in the type and number of ple covariables in a Japanese cohort [29].
variables. One clustering approach, using high-dimensional
electronic medical records and extensive genotype data, iden- Subgroup variables The clinical relevance of the novel
tified three subtypes of type 2 diabetes enriched in CVD, subgroups has been assessed in multiple cohorts including
nephropathy, retinopathy, neurological diseases and cancer ethnically diverse populations that lack some of the aforemen-
[21]. A second approach used data for 94 type 2 diabetes- tioned clustering variables (most often C-peptide measure-
associated gene variants and 47 diabetes-related traits to ments). Partial replication of the subgroup classification and
subgroup genetic loci according to mechanistic pathways differential risk of complications was reported in cohorts from
and to relate the clinical characteristics of people with type 2 Europe [34, 35], the USA [36, 37], Mexico [38], Latin
diabetes to their genetic risk scores [22]. This study found two America and the Caribbean [39], India [40] and China [41]
clusters of genetic loci related to insulin deficiency and three and in large international trial populations (DEVOTE/
Diabetologia (2022) 65:1770–1781 1773

Table 1 Metabolic characteristics

and diabetes-related complica- Diabetes Metabolic Diabetes-related complications
tions of individuals in the novel subgroup characteristics
diabetes subgroups
SAID Early-onset diabetes Ketoacidosis at diagnosis [6]
Low BMI High risk of retinopathy [29]
High HbA1c High incidence of CKD but dependent on baseline eGFR [28]
Insulin deficiency
Presence of GADA
SIDD Early-onset diabetes Ketoacidosis at diagnosis [6]
Low BMI High risk of retinopathy [6]
High HbA1c Highest prevalence of DSPN [26]
Insulin deficiency Highest prevalence of CAN [26]
GADA negative High prevalence of erectile dysfunction [33]
SIRD Late-onset diabetes Highest liver fat content, fatty liver index, NAFLD fibrosis score and
High BMI prevalence of NAFLD [6, 26, 29]
Most Highest risk for macroalbuminuria, CKD and end-stage renal disease
insulin-resistant [6, 26, 29]
GADA negative High risk of coronary event and stroke (dependent on age and sex) [6]
High prevalence of erectile dysfunction [33]
MOD Early-onset diabetes Intermediate prevalence and risk of diabetes-related complications [6,
High BMI 26]
Intermediate insulin
resistance
GADA negative
MARD Late-onset diabetes High risk of coronary events and stroke (dependent on age and sex) [6]
Low BMI
GADA negative

Metabolic characteristics are based on European cohorts with newly diagnosed diabetes using GAD antibodies,
age at diagnosis, BMI at diagnosis, HbA1c and HOMA-2 estimates of insulin resistance and beta cell function
calculated from fasting glucose and fasting C-peptide concentrations as clustering variables [6, 26]
GADA, GAD antibodies

LEADER/SUSTAIN-6 [42]) (Table 2). At present it is not SIRD and MOD [29]. The association of circulating triacyl-
known whether C-peptide or insulin are required to identify glycerols and inflammatory processes with insulin resistance
SIRD, so it would be informative to compare different combi- is in line with the uniform mechanism underlying common
nations of clustering variables (e.g. with and without C- insulin resistance in humans [45]. A multimarker approach
peptide or insulin) in the same cohorts, to better understand in the GDS found that 23 biomarkers of inflammation differed
their relevance for reclassification. between the subgroups, with biomarker levels in general being
Only a few studies have explored biomarkers and pathways highest in SIRD and lowest in SIDD [46]. After adjustment
underlying differences between subgroups that could deter- for the clustering variables, serum caspase-8 (CASP-8), S100
mine susceptibility to diabetes-related complications. Given calcium-binding protein A12 (EN-RAGE) and IL-6 showed at
the role of lipid metabolism in diabetes, it is noteworthy that least one pairwise difference between the subgroups. The
serum triacylglycerol levels were found to be highest and association between inflammation and insulin resistance
HDL-cholesterol levels lowest in SIRD, while there were no reflects the contribution of inflammation-related processes to
differences in total or LDL-cholesterol [26, 29, 43]. SIRD, whereas inflammatory processes appear less relevant in
Circulating levels of angiopoietin-like protein 8 SIDD [46]. A second study in this cohort showed that the
(ANGPTL8), a regulator of lipid metabolism, were higher in SIRD subgroup also had high leucocyte numbers and the
SIDD, SIRD and MARD than in MOD [44]. However, these highest CD4+ T cell percentages, thereby demonstrating
differences were not adjusted for the clustering variables. different immune cell frequencies between subgroups and
highlighting the proinflammatory characteristics of SIRD
Subgroup differences in inflammation Circulating levels of [47]. Of note, studies on autoimmune diabetes identified both
high-sensitivity C-reactive protein (hsCRP) were highest in genetic and epigenetic determinants of T cell function, with
1774 Diabetologia (2022) 65:1770–1781

Table 2 Overview of clustering studies using alternative demographic and clinical variables to identify subgroups of diabetes

Cohort characteristic Clustering variables Subgroups Specific findings Ref.

VNDS, Italy Age, BMI, HOMA-2 estimates of SIDD Replication of SIDD and MARD [34]
(739 with T2D) beta cell function and insulin MARD OIRD comprising MOD and SIRD
resistance OIRD MARD associated with CVD
EOD Highest HbA1c after 14-month
follow-up in SIDD
Three cohort studies from Age, BMI, HbA1c, random or fasting Five distinct T2D subgroups Three subgroups could be mapped [35]
Europe: Hoorn DCS; C-peptide, HDL-cholesterol back to the original ANDIS
GoDARTS; ANDIS clusters (SIDD, SIRD, MOD)
(15,940 people with T2D, Two subgroups (MD and MDH
within 2 years of related to MARD)
diagnosis) Progression to insulin fastest for
SIDD and slowest for MDH
MASALA and MESA Age at diagnosis, BMI Five T2D subgroups: older age, Older age most common subgroup [36]
multi-ethnic cohorts HbA1c, HOMA estimates of beta cell severe hyperglycaemia, severe for all race/ethnicities apart from
from USA function and insulin resistance obesity, younger age at onset; South Asians
(1293 people with requiring insulin medication use Severe hyperglycaemia subgroup
diabetes; mean most frequent in South Asians
diabetes duration Risk for renal complications and
5.7 years) subclinical CVD differed by
subgroup and by race/ethnicity
Look AHEAD Age at diagnosis, BMI, WC, HbA1c Four subgroups: by older age at Interaction between lifestyle [37]
(5145 overweight/obese diabetes onset; poor glucose intervention and diabetes
people with T2D and control; severe obesity; younger subgroups for three composite
10 years of lifestyle age at diabetes onset cardiovascular outcomes
intervention or control Increased cardiovascular risk for
group) people in subgroup with poor
glucose control randomised to
lifestyle intervention
NHANES (USA) and Models based on different Four subgroups: obesity-related; Risk of retinopathy highest for [38]
four Mexican cohorts combinations of years since insulin-deficient; insulin-resistant; insulin-deficient subgroup and
(1758 people with T2D in diagnosis, BMI, HbA1c, age-related lowest for obesity-related
NHANES; 9887 HOMA-2 estimates of beta cell subgroup
people with T2D in the function and insulin resistance, Subgroup transitions observed after
open-population fasting plasma glucose, 3 months, 1 year and 2 years
Mexican cohorts) METS-IR, METS-VF, age at
diabetes onset
Thirteen cohort studies Age, sex, BMI, WC, Four clusters: Cluster 0, highest BP; Heterogeneous distribution of [39]
from nine countries in systolic/diastolic BP, T2D family Cluster 1, highest BMI and WC, clusters across countries
Latin America and the history highest proportion of positive
Caribbean family history of diabetes; Cluster
(8361 people with T2D) 2, most beneficial risk profile;
Cluster 3, highest age
Electronic medical Age at diagnosis, BMI, WC, HbA1c, Four clusters: Cluster 1, SIDD; SIDD and MARD similar to [40]
records of a tertiary triacylglycerols, Cluster 2, IROD; Cluster 3, diabetes subgroups in other
diabetes centre, India HDL-cholesterol, C-peptide CIRDD; Cluster 4, MARD populations
(19,804 people with T2D; (fasting and stimulated) IROD and CIRDD unique to Asian
diabetes duration Indian population
<5 years) IROD showed highest BMI and
highest C-peptide levels
CIRRD showed lowest age of onset,
highest serum triacylglycerols,
highest risk for kidney disease
Retrospective Age at diagnosis, BMI, HbA1c, Replication of SAID, SIRD and Higher risk for retinopathy, [41]
clinic-based study HOMA-2 estimates of beta cell MARD when using the original peripheral neuropathy,
sample, PR China function and insulin resistance, six clustering variables hypertension and CKD for SIRD
(5414 people with T2D; GADA; additional model with Replication of SAID, SIDD, SIRD, (vs IRD)
mean diabetes duration triacylglycerols and uric acid MOD and MARD and Higher risk for retinopathy and
8.6 years) identification of novel subgroups diabetic foot for SIDD (vs IRD)
(UARD, IRD) when all clustering
variables were used
Diabetologia (2022) 65:1770–1781 1775

Table 2 (continued)

Cohort characteristic Clustering variables Subgroups Specific findings Ref.

Three global Age at diagnosis, BMI, HbA1c Identification of four subgroups: Differences between clusters for [42]
cardiovascular clusters A–D major adverse cardiovascular
outcomes trials: events, cardiovascular death,
DEVOTE, LEADER, nephropathy and severe
SUSTAIN-6 hypoglycaemia when comparing
(20,274 people with T2D; subgroups in at least one cohort
follow-up of
2.0–3.8 years)

CIRDD, combined insulin-resistant and deficient diabetes; DCS, Diabetes Care System; DEVOTE, Trial Comparing Cardiovascular Safety of Insulin
Degludec vs Insulin Glargine in Patients With Type 2 Diabetes at High Risk of Cardiovascular Events; EOD, early-onset diabetes; GADA, GAD
antibodies; GoDARTS, Genetics of Diabetes Audit and Research; IRD, inheritance-related diabetes; IROD, insulin-resistant obese diabetes; LEADER,
Liraglutide Effect and Action in Diabetes: Evaluation of Cardiovascular Outcome Results; MASALA, Mediators of Atherosclerosis in South Asians
Living in America; MD, mild diabetes; MDH, mild diabetes with high cholesterol; MESA, Multi-Ethnic Study of Atherosclerosis; METS-IR, Metabolic
score for insulin resistance; METS-VF, metabolic score for visceral fat; NHANES, National Health and Nutrition Examination Survey; OIRD, obese
insulin-resistant diabetes; SUSTAIN-6, Trial to Evaluate Cardiovascular and Other Long-term Outcomes With Semaglutide in Subjects With Type 2
Diabetes; T2D, type 2 diabetes; UARD, uric acid-related diabetes; WC, waist circumference

effects on gene expression [48, 49]. (Epi)genetic variation and Subgroups of individuals with different risk
its impact on transcriptomes in immune cells will require more of progression to type 2 diabetes
detailed analyses with respect to relevance in disease aetiolo- and complications
gy in the other subgroups.
Trajectory analyses show that changes in metabolic and
Genetic predisposition There is evidence that the subgroups inflammation-related biomarkers start >10 years before the
may differ in their associations with gene variants predispos- onset of type 2 diabetes [51–53]. Therefore, it is also of high
ing to diabetes. The HLA SNP rs2854275 showed the same clinical relevance to identify subgroups of individuals at
association with SAID as with type 1 diabetes in previous different risk for diabetes and for complications, which may
studies, but not with SIDD, pointing towards different aetiol- start even before the manifestation of diabetes. Latent-class
ogies of insulin deficiency in the subgroups [6]. Both the trajectory analysis using OGTTs in people without diabetes
TCF7L2 SNP rs7903146, which is known for its association revealed four subgroups that differed in anthropometric, meta-
with type 2 diabetes, and a genetic score for type 2 diabetes bolic and inflammation-related variables [24], but this study
were associated with SIDD, MOD and MARD but not with did not analyse diabetes-related complications.
SIRD [6]. A genetic risk score for insulin secretion was asso- A recent study in a cohort of individuals at elevated risk of
ciated with MOD and MARD (and nominally with SIDD) but type 2 diabetes explored the pathophysiological heterogeneity
again not with SIRD [6]. This suggests a more pronounced before clinical diabetes onset [23]. Participants from the
role for genetic predisposition to SIDD, MOD and MARD Tübingen Family Study (TUEF) and Tübingen Lifestyle
and a stronger role for environmental determinants in SIRD. Intervention Program (TULIP) underwent clustering based on
Of note, individuals with SIRD were more frequently carriers OGTT, MRI (body fat distribution, liver fat), serum lipids and
of the G allele of rs738409 in PNPLA3, the gene encoding a PRS for type 2 diabetes. This study found six subphenotypes
patatin-like phospholipase domain-containing-3, which is differing in diabetes-related variables: 1, low risk; 2, very low
characterised by its positive association with hepatic fat risk; 3, beta cell failure; 4, low-risk obese; 5, high-risk insulin-
content [43] and may contribute to the relationship between resistant fatty liver; 6, high-risk visceral fat nephropathy [23].
SIRD and progression of NAFLD to fibrosis [6, 26, 29]. Thus, Results were replicated in the Whitehall II cohort using a reduced
genetic analyses corroborate the difference between SAID and set of clustering variables. Overall, clusters 3, 5 and 6 showed
the other subgroups but suggest unique mechanisms that higher glucose levels at baseline but only clusters 3 and 5 had an
might distinguish SIRD from SIDD, MOD and MARD. increased incidence of type 2 diabetes. Clusters 3, 5 and 6
At present, the only study integrating genetic, featured the highest CKD risk and higher intima–media thick-
metabolomic, lipidomic and proteomic data to compare diabe- ness, and clusters 5 and 6 had the highest all-cause mortality.
tes subtypes was based on different clustering variables (age, Data from Whitehall II indicated that individuals from the low-
BMI, HbA1c, HDL-cholesterol, and random or fasting C- risk clusters 1, 2 and 4 transitioned to MOD and MARD with
peptide). SIRD showed the most distinct molecular signature, diabetes onset, whereas individuals from the high-risk cluster 6
mostly related to insulin resistance, lipids and inflammation transitioned to SIRD [23]. Thus, clustering approaches can also
[50]. identify subphenotypes with respect to glycaemic, renal,
1776 Diabetologia (2022) 65:1770–1781

cardiovascular and all-cause mortality risk, corresponding to insulin resistance (e.g. by future insulin sensitisers) (Fig. 1).
previous findings for overt diabetes [6, 26]. Given the proinflammatory profile and the high risk of compli-
cations in SIRD, novel therapies targeting inflammatory path-
ways, as developed for people at high cardiovascular risk [54,
Translation into clinical practice: therapeutic 55], could also be considered in the future. Initially, lifestyle
implications modification and metformin are sufficient for treating MOD
and MARD. Nevertheless, MOD may specifically benefit from
RCTs are required to evaluate the clinical relevance of reclassi- weight loss intervention by hypo-energetic diets and drugs,
fication efforts. Until data from subgroup-specific RCTs are whereas MARD may be better treated by nutrition avoiding
available, it is only possible to investigate in cohort studies [6, further ageing-related sarcopenia (Fig. 1).
29] or intervention trials, such as A Diabetes Outcome Currently, it is not clear whether our knowledge on mech-
Progressive Trial (ADOPT) [28], whether individuals allocated anisms (and adverse effects) of these drugs will translate into
to subgroups differ in their treatment at study baseline or their subgroup-specific treatment benefits [56]. However, the large
treatment responses, respectively (Table 3). Importantly, the high number of ongoing RCTs using novel therapeutic agents
frequency of individuals without initial glucose-lowering treat- targeting insulin secretion, insulin resistance, liver metabolism
ment, the low frequency of insulin use and the shortest time to and other mechanisms that differ between the subgroups holds
reach the HbA1c target were similar for SIRD, MOD and MARD promise for precision healthcare [57].
and correspond to their less pronounced insulin deficiency So far, only the Look AHEAD (Action for Health in Diabetes)
compared with SAID and SIDD (Table 3). Glycaemic deteriora- study analysed the differential response to lifestyle intervention
tion may thus suggest a milder progression of disease for SIRD, [37]. Individuals with type 2 diabetes were allocated to four
MOD and MARD. However, the higher risk for several compli- subgroups, which are not directly comparable with the previously
cations in SIRD clearly indicates the need for treatment intensi- described subgroups [6, 26] due to differences in clustering vari-
fication addressing CKD, CVD and NAFLD (e.g. by sodium– ables. Randomisation to intensive lifestyle intervention was asso-
glucose cotransporter 2 inhibitors [SGLT2is] and glucagon-like ciated with increased cardiovascular risk in the subgroup
peptide-1 receptor agonists [GLP-1RAs]), as well as targeting characterised by the poorest glucose control and most frequent

Table 3 Novel diabetes subgroups: glucose-lowering therapy in cohort studies and response to therapy in ADOPT

Diabetes Therapy in cohort studies Response to therapy in ADOPT Comment

subgroup

SAID Most frequent use of insulin and lowest use of Not analysed in the context of novel diabetes Findings are in line with the established
metformin at baseline [6, 29] subgroups treatment for type 1 diabetes and
Shortest time to sustained insulin use [6] LADA
SIDD Most frequent use of metformin at baseline Initial treatment response best with Data are in line with the low beta cell
[8, 29] sulfonylureas but highest HbA1c increase reserves in this subgroup
Frequent use of insulin at baseline and short time thereafter with sulfonylureas
to sustained insulin use, although less
pronounced than for SAID [6]
Shortest time to treatment with oral medication
other than metformin and longest time to reach
HbA1c treatment goal [6]
SIRD Most frequently treated with metformin or HbA1c benefit with thiazolidinedione therapy Findings are plausible given the
without glucose-lowering therapy [6] pronounced insulin resistance and high
Evidence for higher insulin use later after prevalence of NAFLD in SIRD
diabetes diagnosis [29]
MOD Most frequently treated with metformin or Initial treatment response best with Data indicate a mild form and mild
without glucose-lowering therapy [6] sulfonylureas but highest HbA1c increase progression of diabetes
Lowest baseline use of insulin [29] thereafter with sulfonylureas
MARD Most frequently treated with metformin or HbA1c benefit with sulfonylurea therapy, Data indicate a mild form and mild
without glucose-lowering therapy [6] limited to about 2 years, vs metformin and progression of diabetes
Low cumulative incidence of treatment with oral thiazolidinedione treatment
medication other than metformin or of
sustained insulin use [6]

Data for response to therapy in ADOPT are from a secondary analysis of the trial [28], which randomised newly diagnosed, drug-naive individuals with
type 2 diabetes to metformin, sulfonylurea (glibenclamide) or thiazolidinedione (rosiglitazone) monotherapy
LADA, latent autoimmune diabetes of adults
Diabetologia (2022) 65:1770–1781 1777

Gaps in the current knowledge, open questions and future

research directions
• The proposed novel diabetes subgroups [6] are partly based on fasting C-peptide, which is not always
available, so the selection of clustering variables should be optimised to ensure simple and reproducible
clinical measures
• The original cluster algorithm assessed autoimmunity only by measuring GADA [6]; adding other islet-
directed antibodies will expand the subgroup with autoimmune diabetes [26]
• Some people with diabetes migrate between diabetes subgroups over time [26, 38]. More studies in-
vestigating this phenomenon and linking it to disease progression are warranted to assess the stability
of subgroup allocation and its impact on the utility of this concept
• Different strategies have been proposed to reclassify diabetes [58, 59]. Future studies should compare
the clinical utility of distinct diabetes subgroups and probabilistic models using continuous risk factors
for the prediction of complications and treatment response. Irrespective of the underlying statistical
method, the practicability in clinical care of different treatment decisions depending on reclassification
needs to be proven
• Both phenotypic and genotypic data could be used to reclassify diabetes [6, 22]. Given the wide age
range of diabetes onset across the lifespan both approaches may be complementary but with different
relative contributions to risk prediction depending on age
• Biomarker studies (e.g. on biomarkers of inflammation) point towards pathomechanisms that differ be-
tween diabetes subgroups [43, 46, 47]. An extension of phenotyping including not only multiple omics
data (e.g. genomics, metabolomics, proteomics, transcriptomics) but also deep molecular and physio-
logical phenotyping using wearable devices may help to provide further insight and to refine the subtyp-
ing of diabetes [10, 60-62]
• Individual responses to diet and exercise vary between individuals [63, 64]. It is not known to what
extent this individuality is also related to different pathomechanisms that characterise the novel diabetes
subgroups
• The five diabetes subgroups were developed using data from European cohorts [6] but diabetes phe-
notypes and drug responses can differ between ethnic groups [59, 65, 66]. Replication studies in non-
White people with diabetes identified additional subgroups but were often based on different sets of
clustering variables and/or diabetes durations [27, 29, 30, 36, 38-41]. Overall, the applicability of the
subgroup concept in the general population and in different ethnic groups requires further study
• Studies have focused so far on associations between diabetes subgroups and objective clinical
measures but also need to take into account patient-centred mental health and quality-of-life outcomes
[13]
• Given the novel findings on subgroups with different risk trajectories among people with and without
diabetes [6, 23], future investigations should assess whether individuals in these subgroups require not
only different intensities in drug treatment or lifestyle intervention programmes but also different intervals
in screening and monitoring for complications
• If it becomes a requirement for the approval of new drugs to demonstrate whether they are equally
effective in different subgroups or just recommended for specific subgroups, smaller market shares for
precision medicines may entail challenges for future drug development [67]

use of insulin [37]. Thus, subgroups may differ in their response Methodological aspects and open questions
to non-pharmacological treatment, emphasising the differential
need for pharmacological treatment intensification to prevent The text box above gives an overview of key gaps in our
diabetes-related complications. current knowledge, resulting open questions, and future
1778 Diabetologia (2022) 65:1770–1781

Subclassification Precision prognostics Precision prevention Precision treatment

Type 2 Type 1

(Early) Insulin
SAID
DPP4i?

(Sulfonylureas?)

SIDD

Endurance
exercise?
Metformin +
Low energy/
SIRD saturated fat diet? SGLT2i? GLP1-RA?

PPARa?

Dual agonists?
Resistance
MOD Anti-inflammatory
exercise?
drugs?
High-protein diet?

MARD

Fig. 1 Possible future implications of precision diabetology based on the sulfonylurea. SIRD and MOD would benefit from medication that
novel diabetes subgroups. Although the utility of the concept needs to be induces weight loss (SGLT2i, GLP-1RA, dual agonist) or also addresses
evaluated in RCTs, one may speculate on the potential implications of a risk of CVD or nephropathy (SGLT2i, GLP-1RA). Providing that safety
new (sub)classification of diabetes for tailored diagnosis, prevention and and efficacy have been established, new insulin sensitisers (e.g. peroxi-
treatment. Individuals in the different diabetes subgroups differ in their some proliferator activator receptor agonists) or anti-inflammatory drugs
susceptibility to developing specific complications. The different (patho- could also improve targeted treatment of SIRD. On the other hand, indi-
physiological) phenotypes may also differ in their response to lifestyle- viduals with MARD should receive treatments avoiding weight loss and
related and pharmacological strategies. SAID requires early introduction sarcopenia (e.g. protein-balanced diets and moderate resistance training).
of insulin supplementation, whereas SIDD may also benefit from a dipep- PPARa, peroxisome proliferator activator receptor agonist. This figure is
tidyl peptidase 4 inhibitor (DPP4i) or, when cost is a major issue, a available as a downloadable slide

directions in this field; some of the methodological aspects An alternative strategy in precision diabetology may be
are also briefly discussed here. Any effort made concern- based on statistical models using continuous risk factors [28,
ing reclassification into subgroups has a strong conceptual 58, 68]. In a secondary analysis of RCTs [28], age at diabetes
appeal because it is easy to communicate and implement diagnosis and renal function at baseline were better predictors
once RCTs have shown subgroup-specific differences to of disease progression than the subgroup assignment according
non-pharmacological and pharmacological interventions. to Ahlqvist et al [6]. Thus, specific phenotypic measures to
However, this approach assumes a certain degree of homo- predict glycaemic progression, onset of complications and treat-
geneity within, and clear differences between, subgroups, ment response could be used to optimise diabetes care in an
whereas in reality the characteristics of individuals from individualised approach. The risk assessment could be updated
different clusters partially overlap [3]. The subgroup regularly to take into account disease progression, with corre-
approach is also limited by the fact that subgroup assign- sponding treatment changes. However, these models would
ment requires the availability of all clustering variables. only be useful for optimising one specific outcome such as
Finally, the utility of subgroups depends on their stability. glycaemic progression or the development of a predefined
An analysis in the GDS demonstrated that 23% of the study complication unless they were a priori designed to predict a
participants migrated into a different subgroup within the composite endpoint comprising different outcomes based on
first 5 years after the diagnosis of diabetes [26]. the patients’ preferences. Currently, such an approach remains
Diabetologia (2022) 65:1770–1781 1779

challenging because it requires a huge amount of individual- but unedited supplementary material including a slide of the figure for
download, which is available to authorised users.
participant data to develop the underlying models.
One general criticism of the aforementioned reclassification Funding Open Access funding enabled and organized by Projekt DEAL.
strategies refers to their use of phenotypic data that depend on The German Diabetes Center (DDZ) is supported by the Ministry of
disease progression, lifestyle and medication and therefore Culture and Science of the State of North Rhine-Westphalia and the
necessitate regular adaptation. In contrast, genotypic data are German Federal Ministry of Health. This work was supported in part
by a grant from the German Federal Ministry of Education and
stable over time and are more likely to be related to causal Research to the German Center for Diabetes Research (DZD).
mechanisms [22]. However, the proportion of diabetes risk that
can be explained by environmental risk factors is still greater Authors’ relationships and activities CH received a research grant from
than the proportion that can be attributed to known genetic risk Sanofi-Aventis outside the submitted work. CH is a member of the edito-
rial board of Diabetologia. MR received fees as a member of advisory
variants. This means that people with large differences in genet- boards or as a speaker from Allergan, Boehringer Ingelheim Pharma,
ic risk scores show minor phenotypic differences that can be Bristol-Myers Squibb, Eli Lilly, Fishawack Group, Gilead Sciences,
overcome by modifying exogenous risk factors [69, 70]. Novartis Pharma, Intercept Pharma, Inventiva, Novo Nordisk, Target
Currently, it is unknown which of the two approaches or alter- NASH and Terra Firma, and has been involved with clinical trial research
for Boehringer Ingelheim, Danone Nutricia Research and Sanofi-Aventis,
natively a combination of both phenotypic and genotypic all outside the submitted work.
reclassification would provide the best benefit.
Irrespective of all methodological and practical issues, it is Contribution statement Both authors were responsible for drafting the
important to emphasise the following points: (1) clinical decision article and revising it critically for important intellectual content. Both
making is always binary at the end (i.e. resulting in the decision authors approved the version to be published.
to treat or not to treat and in the selection of certain non-
pharmacological or pharmacological interventions) and (2) any Open Access This article is licensed under a Creative Commons
approach to reclassify diabetes must result in diabetes prevention Attribution 4.0 International License, which permits use, sharing, adap-
tation, distribution and reproduction in any medium or format, as long as
and care superior to that received under the established classifi-
you give appropriate credit to the original author(s) and the source,
cation. The ongoing initiative of the ADA and the EASD on provide a link to the Creative Commons licence, and indicate if changes
precision medicine in diabetes will provide a detailed roadmap were made. The images or other third party material in this article are
for future studies and application of tailored diagnostics, preven- included in the article's Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in
tion and treatment on the road to precision diabetology [13].
the article's Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a
copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Conclusions

The heterogeneity of diabetes, particularly type 2 diabetes, is

evident from differences in multiple pathophysiological and References
clinical features. Recent studies provided novel insights into
1. Gale EA (2001) The discovery of type 1 diabetes. Diabetes 50(2):
interindividual differences by clustering people with diabetes 217–226. https://doi.org/10.2337/diabetes.50.2.217
into five subgroups, which are reproducible and associated 2. Gale EA (2013) Is type 2 diabetes a category error? Lancet 381(9881):
with different trajectories in disease progression and onset of 1956–1957. https://doi.org/10.1016/S0140-6736(12)62207-7
diabetes-related complications including CKD, retinopathy, 3. McCarthy MI (2017) Painting a new picture of personalised medi-
CVD, NAFLD and neuropathies. Based on the current cine for diabetes. Diabetologia 60(5):793–799. https://doi.org/10.
1007/s00125-017-4210-x
evidence, it is possible to propose innovative stratified preven- 4. Pearson ER (2019) Type 2 diabetes: a multifaceted disease.
tion and treatment approaches at least for some of these Diabetologia 62(7):1107–1112. https://doi.org/10.1007/s00125-
subgroups (Fig. 1). However, the ultimate test of the utility 019-4909-y
of precision diabetology will require RCTs to demonstrate 5. Szendroedi J, Saxena A, Weber KS et al (2016) Cohort profile: the
German diabetes study (GDS). Cardiovasc Diabetol 15:59. https://
whether the probability-based assignment to subgroups and
doi.org/10.1186/s12933-016-0374-9
subsequent subgroup-specific prevention or treatment is 6. Ahlqvist E, Storm P, Käräjämäki A et al (2018) Novel subgroups of
indeed superior to that proposed by the current guidelines. adult-onset diabetes and their association with outcomes: a data-
Furthermore, future studies should address methodological driven cluster analysis of six variables. Lancet Diabetes Endocrinol
issues, in particular on the best precision diabetology 6(5):361–369. https://doi.org/10.1016/S2213-8587(18)30051-2
7. Davies MJ, D’Alessio DA, Fradkin J et al (2018) Management of
approaches, and also uncertainties regarding the transethnic
hyperglycaemia in type 2 diabetes, 2018. A consensus report by the
generalisability of the current findings. American Diabetes Association (ADA) and the European
Association for the Study of diabetes (EASD). Diabetologia
Electronic supplementary material The online version of this article 61(12):2461–2498. https://doi.org/10.1007/s00125-018-4729-5
(https://doi.org/10.1007/s00125-021-05625-x) contains peer-reviewed Erratum in: Diabetologia 2019;62(5):873
1780 Diabetologia (2022) 65:1770–1781

8. Redondo MJ, Hagopian WA, Oram R et al (2020) The clinical study. Diabetes Care 41(8):1740–1748. https://doi.org/10.2337/
consequences of heterogeneity within and between different diabe- dc18-0279
tes types. Diabetologia 63(10):2040–2048. https://doi.org/10.1007/ 25. Obura M, Beulens JWJ, Slieker R et al (2020) Post-load glucose
s00125-020-05211-7 subgroups and associated metabolic traits in individuals with type 2
9. Bonora E, Trombetta M, Dauriz M et al (2020) Chronic complica- diabetes: an IMI-DIRECT study. PLoS One 15(11):e0242360.
tions in patients with newly diagnosed type 2 diabetes: prevalence https://doi.org/10.1371/journal.pone.0242360
and related metabolic and clinical features: the Verona newly diag- 26. Zaharia OP, Strassburger K, Strom A et al (2019) Risk of diabetes-
nosed type 2 diabetes study (VNDS) 9. BMJ Open Diabetes Res associated diseases in subgroups of patients with recent-onset
Care 8(1):e001549. https://doi.org/10.1136/bmjdrc-2020-001549 diabetes: a 5-year follow-up study. Lancet Diabetes Endocrinol
10. Prasad RB, Groop L (2019) Precision medicine in type 2 diabetes. J 7(9):684–694. https://doi.org/10.1016/S2213-8587(19)30187-1
Intern Med 285(1):40–48. https://doi.org/10.1111/joim.12859 27. Zou X, Zhou X, Zhu Z, Ji L (2019) Novel subgroups of patients
11. Gloyn AL, Drucker DJ (2018) Precision medicine in the manage- with adult-onset diabetes in Chinese and US populations. Lancet
ment of type 2 diabetes. Lancet Diabetes Endocrinol 6(11):891– Diabetes Endocrinol 7(1):9–11. https://doi.org/10.1016/S2213-
900. https://doi.org/10.1016/S2213-8587(18)30052-4 8587(18)30316-4
12. Barroso I, McCarthy MI (2019) The genetic basis of metabolic disease. 28. Dennis JM, Shields BM, Henley WE, Jones AG, Hattersley AT
Cell 177(1):146–161. https://doi.org/10.1016/j.cell.2019.02.024 (2019) Disease progression and treatment response in data-driven
13. Chung WK, Erion K, Florez JC et al (2020) Precision medicine in subgroups of type 2 diabetes compared with models based on
diabetes: a consensus report from the American Diabetes simple clinical features: an analysis using clinical trial data.
Association (ADA) and the European Association for the Study Lancet Diabetes Endocrinol 7(6):442–451. https://doi.org/10.
of diabetes (EASD). Diabetologia 63(9):1671–1693. https://doi. 1016/S2213-8587(19)30087-7
org/10.1007/s00125-020-05181-w 29. Tanabe H, Saito H, Kudo A et al (2020) Factors associated with risk
14. Battaglia M, Ahmed S, Anderson MS et al (2020) Introducing the of diabetic complications in novel cluster-based diabetes
Endotype concept to address the challenge of disease heterogeneity subgroups: a Japanese retrospective cohort study. J Clin Med
in type 1 diabetes. Diabetes Care 43(1):5–12. https://doi.org/10. 9(7):2083. https://doi.org/10.3390/jcm9072083
2337/dc19-0880 30. Bennet L, Nilsson C, Mansour-Aly D, Christensson A, Groop L
(2020) Ahlqvist E (2020) adult-onset diabetes in middle eastern
15. Zoppini G, Cacciatori V, Raimondo D et al (2015) Prevalence of
immigrants to Sweden: novel subgroups and diabetic
cardiovascular autonomic neuropathy in a cohort of patients with
complications-the all new diabetes in Scania cohort diabetic
newly diagnosed type 2 diabetes: the Verona newly diagnosed type
complications and ethnicity. Diabetes Metab Res Rev 29:e3419.
2 diabetes study (VNDS). Diabetes Care 38(8):1487–1493. https://
https://doi.org/10.1002/dmrr.3419
doi.org/10.2337/dc15-0081
31. Fedotkina O, Sulaieva O, Ozgumus T et al (2021) Novel reclassi-
16. Zaharia OP, Bobrov P, Strassburger K et al (2018) Metabolic char-
fication of adult diabetes is useful to distinguish stages of β-cell
acteristics of recently diagnosed adult-onset autoimmune diabetes
function linked to the risk of vascular complications: the DOLCE
mellitus. J Clin Endocrinol Metab 103(2):429–437. https://doi.org/
study from northern Ukraine. Front Genet 12:637945. https://doi.
10.1210/jc.2017-01706
org/10.3389/fgene.2021.637945
17. Maalmi H, Herder C, Strassburger K et al (2020) Biomarkers of
32. Xing L, Peng F, Liang Q et al (2021) Clinical characteristics and
inflammation and glomerular filtration rate in individuals with
risk of diabetic complications in data-driven clusters among type 2
recent-onset type 1 and type 2 diabetes. J Clin Endocrinol Metab
diabetes. Front Endocrinol (Lausanne) 12:617628. https://doi.org/
105(12):e4370–e4381. https://doi.org/10.1210/clinem/dgaa622
10.3389/fendo.2021.617628
18. Zoppini G, Trombetta M, Pastore I et al (2021) Glomerular filtra- 33. Maalmi H, Herder C, Bönhof GJ et al (2021) Differences in the
tion rate decline in T2DM following diagnosis. The Verona newly prevalence of erectile dysfunction between novel subgroups of
diagnosed diabetes Study-12. Diabetes Res Clin Pract 22:108778. recent-onset diabetes. Diabetologia. https://doi.org/10.1007/
https://doi.org/10.1016/j.diabres.2021.108778 s00125-21-05607-z (in press)
19. Mahajan A, Taliun D, Thurner M et al (2018) Fine-mapping type 2 34. Bonadonna RC, Trombetta M, Zusi C, Dauriz M, Boselli ML,
diabetes loci to single-variant resolution using high-density impu- Bonora E (2019) Novel subgroups of type 2 diabetes at diagnosis
tation and islet-specific epigenome maps. Nat Genet 50(11):1505– in a southern European cohort. Diabetes 68(Supplement 1):1502-P
1513. https://doi.org/10.1038/s41588-018-0241-6 (abstract). https://doi.org/10.2337/db19-1502-P
20. Sarría-Santamera A, Orazumbekova B, Maulenkul T, Gaipov A, 35. Slieker RC, Donnelly LA, Fitipaldi H et al (2021) Replication and
Atageldiyeva K (2020) The identification of diabetes mellitus cross-validation of type 2 diabetes subtypes based on clinical vari-
subtypes applying cluster analysis techniques: a systematic review. ables: an IMI-RHAPSODY study. Diabetologia 64(9):1982–1989.
Int J Environ Res Public Health 17(24):9523. https://doi.org/10. https://doi.org/10.1007/s00125-021-05490-8
3390/ijerph17249523 36. Bancks MP, Bertoni AG, Carnethon M et al (2021) Association of
21. Li L, Cheng WY, Glicksberg BS et al (2015) Identification of type 2 Diabetes Subgroups with Race/ethnicity, risk factor burden and
diabetes subgroups through topological analysis of patient similar- complications: the MASALA and MESA studies. J Clin Endocrinol
ity. Sci Transl Med 7(311):311ra174. https://doi.org/10.1126/ Metab 106(5):e2106–e2115. https://doi.org/10.1210/clinem/dgaa962
scitranslmed.aaa9364 37. Bancks MP, Chen H, Balasubramanyam A et al (2021) Type 2
22. Udler MS, Kim J, von Grotthuss M et al (2018) Type 2 diabetes diabetes subgroups, risk for complications, and differential effects
genetic loci informed by multi-trait associations point to disease due to an intensive lifestyle intervention. Diabetes Care 44(5):
mechanisms and subtypes: a soft clustering analysis. PLoS Med 1203–1210. https://doi.org/10.2337/dc20-2372
15(9):e1002654. https://doi.org/10.1371/journal.pmed.1002654 38. Bello-Chavolla OY, Bahena-López JP, Vargas-Vázquez A et al
23. Wagner R, Heni M, Tabák AG et al (2021) Pathophysiology-based (2020) Clinical characterization of data-driven diabetes subgroups
subphenotyping of individuals at elevated risk for type 2 diabetes. in Mexicans using a reproducible machine learning approach. BMJ
Nat Med 27(1):49–57. https://doi.org/10.1038/s41591-020-1116-9 Open Diabetes Res Care 8(1):e001550. https://doi.org/10.1136/
24. Hulman A, Witte DR, Vistisen D et al (2018) Pathophysiological bmjdrc-2020-001550
characteristics underlying different glucose response curves: a 39. Carrillo-Larco RM, Castillo-Cara M, Anza-Ramirez C, Bernabé-Ortiz
latent class trajectory analysis from the prospective EGIR-RISC A (2021) Clusters of people with type 2 diabetes in the general
Diabetologia (2022) 65:1770–1781 1781

population: unsupervised machine learning approach using national 55. Lawler PR, Bhatt DL, Godoy LC et al (2021) Targeting cardiovas-
surveys in Latin America and the Caribbean. BMJ Open Diabetes cular inflammation: next steps in clinical translation. Eur Heart J
Res Care 9(1):e001889. https://doi.org/10.1136/bmjdrc-2020-001889 42(1):113–131. https://doi.org/10.1093/eurheartj/ehaa099
40. Anjana RM, Baskar V, Nair ATN et al (2020) Novel subgroups of 56. Veelen A, Erazo-Tapia E, Oscarsson J, Schrauwen P (2021) Type 2
type 2 diabetes and their association with microvascular outcomes diabetes subgroups and potential medication strategies in relation to
in an Asian Indian population: a data-driven cluster analysis: the effects on insulin resistance and beta-cell function: a step toward
INSPIRED study. BMJ Open Diabetes Res Care 8(1):e001506. personalised diabetes treatment? Mol Metab 46:101158. https://doi.
https://doi.org/10.1136/bmjdrc-2020-001506 org/10.1016/j.molmet.2020.101158
41. Xiong XF, Yang Y, Wei L, Xiao Y, Li L, Sun L (2021) Identification 57. Perreault L, Skyler JS, Rosenstock J (2021) Novel therapies with preci-
of two novel subgroups in patients with diabetes mellitus and their sion mechanisms for type 2 diabetes mellitus. Nat Rev Endocrinol
association with clinical outcomes: a two-step cluster analysis. J 17(6):364–377. https://doi.org/10.1038/s41574-021-00489-y
Diabetes Investig 12(8):1346–1358. https://doi.org/10.1111/jdi.13494 58. Dennis JM (2020) Precision medicine in type 2 diabetes: using
42. Kahkoska AR, Geybels MS, Klein KR et al (2020) Validation of individualized prediction models to optimize selection of treatment.
distinct type 2 diabetes clusters and their association with diabetes Diabetes 69(10):2075–2085. https://doi.org/10.2337/dbi20-0002
complications in the DEVOTE, LEADER and SUSTAIN-6 cardio- 59. Ahlqvist E, Prasad RB, Groop L (2020) Subtypes of type 2 diabetes
vascular outcomes trials. Diabetes Obes Metab 22(9):1537–1547. determined from clinical parameters. Diabetes 69(10):2086–2093.
https://doi.org/10.1111/dom.14063 https://doi.org/10.2337/dbi20-0001
43. Zaharia OP, Strassburger K, Knebel B et al (2020) Role of Patatin-
60. Del Prato S (2019) Heterogeneity of diabetes: heralding the era of
like phospholipase domain-containing 3 gene for hepatic lipid
precision medicine. Lancet Diabetes Endocrinol 7(9):659–661.
content and insulin resistance in diabetes. Diabetes Care 43(9):
https://doi.org/10.1016/S2213-8587(19)30218-9
2161–2168. https://doi.org/10.2337/dc20-0329
44. Zou H, Duan W, Zhang Z, Chen X, Lu P, Yu X (2020) The circu- 61. Schüssler-Fiorenza Rose SM, Contrepois K, Moneghetti KJ et al
lating ANGPTL8 levels show differences among novel subgroups (2019) A longitudinal big data approach for precision health. Nat
of adult patients with diabetes and are associated with mortality in Med 25(5):792–804. https://doi.org/10.1038/s41591-019-0414-6
the subsequent 5 years. Sci Rep 10(1):12859. https://doi.org/10. 62. Khoshnejat M, Kavousi K, Banaei-Moghaddam AM, Moosavi-
1038/s41598-020-69091-y Movahedi AA (2020) Unraveling the molecular heterogeneity in
45. Roden M, Shulman GI (2019) The integrative biology of type 2 type 2 diabetes: a potential subtype discovery followed by metabol-
diabetes. Nature 576(7785):51–60. https://doi.org/10.1038/s41586- ic modeling. BMC Med Genet 13(1):119. https://doi.org/10.1186/
019-1797-8 s12920-020-00767-0
46. Herder C, Maalmi H, Strassburger K et al (2021) Differences in 63. Berry SE, Valdes AM, Drew DA et al (2020) Human postprandial
biomarkers of inflammation between novel subgroups of recent- responses to food and potential for precision nutrition. Nat Med
onset diabetes. Diabetes 70(5):1198–1208. https://doi.org/10. 26(6):964–973. https://doi.org/10.1038/s41591-020-0934-0
2337/db20-1054 64. Contrepois K, Wu S, Moneghetti KJ et al (2020) Molecular chore-
47. Ratter-Rieck JM, Maalmi H, Trenkamp S et al (2021) Leukocyte ography of acute exercise. Cell 181(5):1112–1130.e16. https://doi.
counts and T cell frequencies differ between novel subgroups of org/10.1016/j.cell.2020.04.043
diabetes and associate with metabolic parameters and biomarkers of 65. Lee MMY, Ghouri N, McGuire DK, Rutter MK, Sattar N (2021)
inflammation. Diabetes 30:db210364. https://doi.org/10.2337/ Meta-analyses of results from randomized outcome trials compar-
db21-0364 ing cardiovascular effects of SGLT2is and GLP-1RAs in Asian
48. Fasolino M, Goldman N, Wang W et al (2020) Genetic variation in versus white patients with and without type 2 diabetes. Diabetes
type 1 diabetes reconfigures the 3D chromatin Organization of T Care 44(5):1236–1241. https://doi.org/10.2337/dc20-3007
Cells and Alters Gene Expression. Immunity 52(2):257–274.e11. 66. Redondo MJ, Balasubramanyam A (2021) Towards an improved
https://doi.org/10.1016/j.immuni.2020.01.003 classification of type 2 diabetes: lessons from research into the
49. Gao P, Uzun Y, He B et al (2019) Risk variants disrupting heterogeneity of a complex disease. J Clin Endocrinol Metab 22:
enhancers of TH1 and TREG cells in type 1 diabetes. Proc Natl dgab545. https://doi.org/10.1210/clinem/dgab545
Acad Sci U S A 116(15):7581–7590. https://doi.org/10.1073/ 67. Fitipaldi H, McCarthy MI, Florez JC, Franks PW (2018) A global
pnas.1815336116 overview of precision medicine in type 2 diabetes. Diabetes 67(10):
50. Slieker RC, Donnelly LA, Fitipaldi H et al (2021) Distinct molec- 1911–1922. https://doi.org/10.2337/dbi17-0045
ular signatures of clinical clusters in people with type 2 diabetes: an 68. Lugner M, Gudbjörnsdottir S, Sattar N et al (2021) Comparison
IMIRHAPSODY study. Diabetes 10:db201281. https://doi.org/10. between data-driven clusters and models based on clinical features
2337/db20-1281 to predict outcomes in type 2 diabetes: nationwide observational
51. Tabák AG, Jokela M, Akbaraly TN, Brunner EJ, Kivimäki M, study. Diabetologia 64(9):1973–1981. https://doi.org/10.1007/
Witte DR (2009) Trajectories of glycaemia, insulin sensitivity, s00125-021-05485-5
and insulin secretion before diagnosis of type 2 diabetes: an analy- 69. Langenberg C, Sharp SJ, Franks PW et al (2014) Gene-lifestyle inter-
sis from the Whitehall II study. Lancet 373(9682):2215–2221. action and type 2 diabetes: the EPIC interact case-cohort study. PLoS
https://doi.org/10.1016/S0140-6736(09)60619-X Med 11(5):e1001647. https://doi.org/10.1371/journal.pmed.1001647
52. Carstensen M, Herder C, Kivimäki M et al (2010) Accelerated increase
70. Mahajan A, Taliun D, Thurner M et al (2018) Fine-mapping type 2
in serum interleukin-1 receptor antagonist starts 6 years before diagno-
diabetes loci to single-variant resolution using high-density impu-
sis of type 2 diabetes: Whitehall II prospective cohort study. Diabetes
tation and islet-specific epigenome maps. Nat Genet 50(11):1505–
59(5):1222–1227. https://doi.org/10.2337/db09-1199
1513. https://doi.org/10.1038/s41588-018-0241-6
53. Tabák AG, Carstensen M, Witte DR et al (2012) Adiponectin trajec-
tories before type 2 diabetes diagnosis: Whitehall II study. Diabetes
Care 35(12):2540–2547. https://doi.org/10.2337/dc11-2263 Publisher’s note Springer Nature remains neutral with regard to jurisdic-
54. Donath MY, Dinarello CA, Mandrup-Poulsen T (2019) Targeting tional claims in published maps and institutional affiliations.
innate immune mediators in type 1 and type 2 diabetes. Nat Rev
Immunol 19(12):734–746. https://doi.org/10.1038/s41577-019-0213-9