Adult-onset diabetes mellitus is a heterogeneous disease with substantial variability across clinical phenotypes [1]. A phenotype-based clustering approach classifies people with diabetes into more granular subtypes [1] and has been replicated in different populations [2]. These novel subtypes comprise severe autoimmune diabetes (SAID), severe insu-lin-deficient diabetes (SIDD), severe insulin-resistant diabe-tes (SIRD), mild obesity-related diabetes (MOD) and mild age-related diabetes (MARD) [1]. They not only differ in their clinical characteristics such as age at diagnosis, HbA1c, body mass index (BMI), β-cell function (from homeostasis model assessment: HOMA2-B) and insulin resistance (from HOMA2-IR), but also in terms of diabetes-related compli-cations [1, 2]. For example, compared to individuals in the MARD cluster, individuals in the SIDD cluster had a higher hazard rate for retinopathy and individuals in the SIRD cluster had a higher hazard rate for chronic kidney disease and steatotic liver disease [1]. While precision diabetes diagnosis holds a promise for precision treatment of adult-onset diabetes, evidence is still needed to show a benefit of cluster-targeted therapy compared to treatments based on current guidelines. One additional obstacle to adopt this method into clinical practice is a relatively complex cluster assignment algorithm (also referred to as “nearest centroid approach” [2]), which is necessary to identify a person’s diabetes subtype. The goal of the project described here was to develop an easy-to-use online tool for the classification of individuals with new-onset diabetes into the clusters [1]. A second goal was to provide a simple graphical measure of cluster similarity, which can help clinicians assess the close-ness of a given individual to the assigned subtype.
To this end, the German Diabetes Center (DDZ) devel-oped the DDZ Diabetes-Cluster-Tool, which was launched in May 2023. This tool is available at https://diabetescalcul ator.ddz.de/diabetescluster-en/ in English and German lang uage. It was developed with the statistical software R ver-sion 4.2.0 [3] and the Shiny web framework. In the menu, the user can directly input all clinical and laboratory data necessary for cluster assignment: presence of antibodies to glutamic acid decarboxylase (GAD), age at diagnosis, BMI, fasting plasma glucose, fasting C-peptide, HbA1c and sex. Note that fasting plasma glucose and C-peptide values are used to compute the HOMA2-B and HOMA2-IR values, which are needed for cluster assignment. For ease of use, the tool supports multiple unit formats for inputting clinical data. For example, fasting plasma glucose can be supplied as either mmol/l or mg/dl. After entering the information, the cluster assignment is computed automatically with the nearest centroid approach using the centroid and normal-ization information from the All-New Diabetes in Scania (ANDIS) cohort [1]. In the output, the user receives two pieces of information, the diabetes subtype and the “degree of similarity” to each of the five subtypes presented as bar chart.
The degree of similarity is computed based on the dis-tance of the individual to each of the five centroids taking into account all of the entered clinical information. For ease of interpretation, the distances are inverted so that the high-est bar corresponds to the shortest distance and thus the best fit. Moreover, the distances are scaled so that they add up to 1, which makes it easier to compare the subtype similarity between individuals. This simple visualization allows the user to identify “borderline cases” who are similar to several clusters, i.e. where there is not one clear subtype to describe the observed phenotype. Note that the degree of similarity to the SAID cluster will always be either 0 or 1, because assignment to this cluster is determined categorically by the absence or presence of GAD-antibodies.
Figure 1 shows a fictional example of an individual and their output of the DDZ Diabetes-Cluster-Tool. The assigned subtype is MOD and the individual’s phenotype seems to match this cluster well, as indicated by the high degree of similarity. Note that this example individual is rather young but highly obese, as is typical for the MOD subtype [1]. The degree of similarity with the other clusters is considerably smaller, which reflects a larger distance to the respective centroids (e.g. MARD). Although current guidelines do not yet advocate the diabetes clusters, the tool could provide information about the pathophysiology for clinicians to dis-cuss the diabetes phenotype of an individual presenting in clinical practice. Moreover, researchers can use this tool to classify individuals and further advance our insights into the underlying differences between the subtypes.
While diabetes subtypes provide an intuitive framework for understanding disease heterogeneity, it is important to keep in mind their limitations. Subclassification into dis-crete clusters always results in a loss of information com-pared to when the corresponding individual clinical features are considered on a continuous scale. While simplification aids interpretation and communication, specific prediction models should utilize the continuous, individual clinical features [4]. An alternative approach proposed by other researchers is to display the heterogeneity of diabetes in a continuous manner using a tree-like graph structure [5]. A further limitation of the clusters is that they were developed using data at diabetes diagnosis and clustering based on data obtained at subsequent time-points of disease course could lead to a change of cluster assignment [2]. Finally, individuals within a cluster may still be relatively heteroge-neous and some individuals may not clearly belong to any of the clusters. For such individuals, who may present clini-cal features overlapping with several subtypes, the discrete cluster assignment will likely be less informative. The DDZ Diabetes-Cluster-Tool includes a graphical measure of clus-ter similarity to help identify such cases.
Overall, the phenotype-based adult-onset diabetes clus-ters have received considerable attention. With the DDZ Diabetes-Cluster-Tool we provide an easy-to-use web-based application for the technical implementation of dia-betes cluster assignments.
Author contributions T.M. developed the tool, wrote the first draft of the manuscript and edited the article. K.P., K.S. and M.B contributed to the development of the tool and reviewed the article. P.Z., O.S. and O.K. contributed to the discussion and reviewed the article. M.R. and R.W. conceived the idea for the tool, contributed to the discussion, re-viewed and edited the article. R.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors gave final approval of this version to be published.
Funding Open Access funding enabled and organized by Projekt DEAL.
The German Diabetes Study was initiated and financed by the Ger-man Diabetes Center (which is funded by the German Federal Min-istry of Health and the Ministry of Culture and Science of the state of North Rhine-Westphalia), the German Diabetes Association, the German Federal Ministry of Education and Research (to the German Center for Diabetes Research, DZD), the German Research Founda-tion (DFG, CRC-SFB 1116) and the Schmutzler Stiftung. The research of MR is further supported by grants from the European Community (HORIZON-HLTH-2022-STAYHLTH-02-01: Panel A) to the INTER-CEPT-T2D consortium and the German Science Foundation (DFG; RTG/GRK 2576 vivid, Project 3 and 493659010 Future4CSPMM). The research of OPZ is supported by grants from the European Foun-dation for the Study of Diabetes (EFSD Rising Star Award) and the German Diabetes Association (DDG Adam Heller Prize). The project has received funding from the program “Profilbildung 2020”, an ini-tiative of the Ministry of Culture and Science of the State of North Rhine-Westphalia (PROFILNRW-2020-107-B). The sole responsibil ity for the content of this publication lies with the authors. The funding sources had no role in study design, data collection, data analysis, data interpretation, or writing of the report.
Code availability The R code for the DDZ Diabetes-Cluster-Tool is available upon reasonable request.
Declarations
Conflict of interest RW reports lecture fees from Novo Nordisk, Sano-fi-Aventis, Boehringer-Ingelheim and Eli Lilly and served on the ad-visory board for Akcea Therapeutics, Daiichi Sankyo, Sanofi-Aventis, Eli Lilly and NovoNordisk. M.R. received fees for consulting or lec-turing from Astra Zeneca, Boehringer- Ingelheim, Echosens, Eli Lilly Madrigal Pharmaceuticals, MSD Sharp & Dohme, Novo Nordisk and Pfizer and performed investigator-initiated research with support from Boehringer Ingelheim, Danone Nutricia Research and Novo Nordisk. KP reports lecture fees from Berlin-Chemie. OK, OS, OPZ, MB, KS and TM report no conflicts of interest.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in 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.o rg/licenses/by/4.0/.
This article is excerpted from the 《Acta Diabetologica》 by Wound World.
