Precision Medicine in Type 2 Diabetes: Clinical Markers of Insulin Resistance Are Associated With Altered Short- and Long-term Glycemic Response to DPP-4 Inhibitor Therapy

Type 2 diabetes is a heterogeneous condition characterized by varying degrees of reduced β-cell function and higher levels of insulin resistance. Most of the 400 million patients worldwide will at some point require glucose-lowering medication (1). Major international treatment guidelines recommend at least four oral treatment options after initial metformin has failed to achieve control, with choice between these informed predominantly by method of administration, overall side effect profile, and cost (2–5).

Individual response to glucose-lowering therapies in type 2 diabetes varies greatly. Identification of clinical phenotypic features or biomarkers robustly associated with glycemic response or other potentially beneficial effects (for example, reduced weight gain or side effects for each therapy) may allow treatment of patients with the agent that is most likely to be effective for them, an approach known as “precision” or “stratified” medicine (6,7). Although much research has focused on identifying genetic or novel biomarker predictors of response, precision diabetes is most likely to be cost effective and have clinical impact using simple inexpensive biomarkers or routinely available clinical phenotypic features (8,9).

Dipeptidyl peptidase 4 (DPP-4) inhibitors are common (20% of U.S. and 27% of U.K. second-line glucose-lowering prescriptions after metformin in 2013) (10,11), well-tolerated (12) oral therapy options recommended in all clinical guidelines (2–5). Beyond baseline HbA_1c and fasting glucose, it is unclear if other factors are associated with glycemic response to DPP-4 inhibitors (13,14). A major mechanism of action of DPP-4 inhibitors is potentiation of β-cell insulin secretion. We aimed to establish if measures of insulin secretion and insulin resistance were associated with short-term glycemic response and long-term durability of response in patients with type 2 diabetes starting DPP-4 inhibitor therapy.

We assessed whether clinical features and biomarkers associated with insulin secretion and insulin resistance were predictive of short-term 6-month glycemic response in analysis of a prospective study of patients starting DPP-4 inhibitor therapy as part of routine care (Predicting Response to Incretin Based Agents [PRIBA] study). To validate our findings, we tested the consistency of associations between routinely recorded factors associated with response in PRIBA in a retrospective analysis of a much larger group of patients from the U.K. Clinical Practice Research Datalink (CPRD), evaluating both 6-month glycemic response and long-term durability of response to 3 years.

The PRIBA study was designed to test the hypothesis that those who have low insulin secretion, as measured by C-peptide, will have poor glycemic response to incretin-based treatments (https://clinicaltrials.gov/ct2/show/NCT01503112), with associations between glycemic response and other clinical features, islet autoantibodies and HOMA2 estimates of β-cell function, and insulin sensitivity evaluated in prespecified secondary analysis. Three hundred five participants due to start DPP-4 inhibitor therapy as part of their usual care were recruited from primary and secondary care across 17 National Institute for Health Research (NIHR) clinical research network centers in the U.K. from April 2011 to October 2013 as previously described (15).

At baseline (immediately prior to starting therapy), we measured HbA_1c, fasting glucose, clinical markers of insulin resistance and insulin secretion (fasting C-peptide and postmeal urine C-peptide creatinine ratio [UCPCR] [16,17], BMI, triglycerides, HDL-cholesterol [HDL-c] [18], sex hormone–binding globulin [SHBG], GAD, and IA2 islet autoantibodies), and other clinical characteristics (age at therapy, sex, duration of diabetes, estimated glomerular filtration rate [eGFR], ethnicity, LDL-cholesterol [LDL-c], and number of diabetes therapies). We calculated HOMA2-%B and HOMA2 estimates of insulin resistance (HOMA2-IR) from fasting glucose and C-peptide measures using the HOMA2 calculator available from http://www.dtu.ox.ac.uk/homacalculator/ (19). Laboratory analysis was conducted as previously reported (15). Participants were included in the analysis if they were not insulin treated and had at least 3 months’ follow-up with >75% adherence to therapy and limited cotreatment change (see study profile in Supplementary Fig. 1A). Ethics approval was granted by the South West National Research Ethics Committee, and all participants gave written informed consent.

CPRD is the world’s largest longitudinal database of anonymized primary care electronic health records (20). We included 23,001 noninsulin-treated patients with type 2 diabetes with prescription records of starting a DPP-4 inhibitor for the first time from June 2007 to September 2016 and followed them up while they remained on DPP-4 inhibitor therapy without the addition or cessation of any other antihyperglycemic medication (see study profile in Supplementary Fig. 1B). We extracted baseline routine clinical characteristics (age at therapy, duration of diabetes, sex, and BMI) and biomarkers (HbA_1c, triglycerides, HDL-c, LDL-c, and eGFR), with baseline defined as the most recent record in the 3 months prior to the drug start date. Ethics approval was granted by the CPRD Independent Scientific Advisory Committee (ISAC 13_177R).

The primary outcome was the absolute change from baseline in HbA_1c 6 months after starting therapy, adjusting for baseline HbA_1c. When a 6-month HbA_1c was not available or eligible in the PRIBA study (Supplementary Fig. 1A), we used a 3-month HbA_1c measure, as previously described (15). In CPRD, a valid 6-month HbA_1c was defined as the closest HbA_1c to 6 months after the drug start date ±3 months for patients on unchanged antihyperglycemic therapy.

In CPRD, in which long-term follow-up data were available, we assessed durability of response as the time to glycemic failure up to 3 years in a complete case analysis of patients with baseline HbA_1c between 53 and 97 mmol/mol (7–11%) and at least 3 months on DPP-4 inhibitor therapy (n = 15,616). Glycemic failure was defined as 1) two consecutive HbA_1c measurements >69 mmol/mol (8.5%) or 2) a single HbA_1c measurement >69 mmol/mol (8.5%) followed by the addition of another antihyperglycemic therapy. To examine the sensitivity of results to this definition, we repeated the analysis using HbA_1c thresholds of 1) 53 mmol/mol (7.5%) and 2) the baseline HbA_1c level specific to each individual patient.

We examined associations between each standardized marker of insulin resistance and insulin secretion and 6-month HbA_1c response in a series of linear regression models adjusted for baseline HbA_1c and, in PRIBA, cotherapy change (13,20). Nonnormally distributed variables were log-transformed. We conducted a complete case analysis for each marker, including all patients with valid data even if they had missing data for other markers. To evaluate model fit, we examined normality of residuals and linearity of associations for continuous variables. In both data sets, we tested the independence of initial associations for each marker of insulin resistance and insulin secretion with 6-month response in further multivariable analysis, controlling for baseline HbA_1c and other routinely recorded characteristics: age at therapy, duration of diabetes, sex, eGFR, LDL-c, ethnicity (CPRD only: white, nonwhite, and missing), and cotherapy change (PRIBA only; CPRD patients all on unchanged therapy).

To further assess the robustness of findings, we repeated the baseline adjusted analysis of 6-month response for males and females separately in both data sets and in PRIBA with additional adjustment for fasting glucose. In CPRD, we repeated the baseline adjusted analysis using 12-month response as the outcome in a distinct cohort of patients with a 12-month (closest ±3 months as for definition of 6-month response) HbA_1c record (n = 16,166).

Based on the initial results, we defined three patient subgroups by standard clinical cutoffs for obesity (BMI ≥30 kg/m²) and high triglycerides (≥2.3 mmol/L) (21): group A, nonobese and normal triglycerides; group B, nonobese or normal triglycerides; and group C, obese and high triglycerides. We estimated the mean 6-month HbA_1c response for each subgroup using linear regression models adjusted for baseline HbA_1c and, in PRIBA, cotherapy change. We standardized baseline HbA_1c to the mean PRIBA baseline level of 74 mmol/mol (8.9%) for all subgroups in both data sets.

For three subgroups defined by the same BMI and triglyceride thresholds, we compared mean durability in response to 3 years after starting therapy using a flexible parametric time to failure survival model. We included all patients with at least 3 months on therapy after starting a DPP-4 inhibitor with valid baseline records of all covariates (baseline HbA_1c, age at therapy, duration of diabetes, sex, and eGFR). The use of flexible parametric models allowed prediction of the probability of therapy failure over 3 years as well as hazard ratios (HRs) consistent with Cox proportional hazards regression (22). We tested continuous variables for nonlinearity and evaluated proportional hazards assumptions using Schoenfeld residuals. To estimate the probability of therapy failure for each subgroup, a predicted survival curve was calculated for each patient in the data set before the individual survival curves for all patients within a subgroup were averaged (23). Each curve was standardized to the mean CPRD values of other clinical covariates (baseline HbA_1c 72 mmol/mol [8.7%]; age at therapy 64 years; duration of diabetes 8 years; and eGFR 82 mL/min/1.73 m²). Point estimates for the failure probability at 3 years by subgroup were calculated using the same approach.

To test the specificity of findings for DPP-4 inhibitors, we repeated the analyses of short-term response and durability of response for noninsulin-treated subjects starting glucagon-like peptide 1 (GLP-1) receptor agonists, the other glucose-lowering drug evaluated in PRIBA (PRIBA, n = 339; CPRD, n = 4,464). We have previously reported the PRIBA primary analysis of predictors of glycemic response for the full PRIBA GLP-1 receptor agonist cohort, which included an additional 209 insulin-treated participants (14). All data extraction and analysis were conducted using Stata v14.0 (StataCorp, College Station, TX).

Baseline characteristics and biomarker measures were similar for subjects starting DPP-4 inhibitors in both data sets (Table 1). In both cohorts, the majority of patients started sitagliptin. Two hundred fifty-four patients were included in PRIBA and 23,001 (for analysis of 6-month glycemic response) in CPRD (for study profiles, see Supplementary Fig. 1). Mean (SD) 6-month HbA_1c change was −8.3 (13.5) mmol/mol (−0.7% [1.2%]) in PRIBA and −7.6 (15.1) mmol/mol (−0.7% [1.4%]) in CPRD.

In the PRIBA cohort, mean HbA_1c response was reduced by 1.67 mmol/mol for every 1 SD higher baseline fasting C-peptide (standardized β 1.67 [95% CI 0.17, 3.17] mmol/mol/SD; P = 0.03) (Fig. 1). We observed the same direction and similar size of effect for UCPCR (response reduction per SD higher 1.65 [95% CI −0.07, 3.37] mmol/mol; P = 0.06). Higher baseline HOMA2-IR was also associated with reduced response (response reduction per SD higher: 2.17 [95% CI 0.62, 3.72] mmol/mol; P = 0.01), but there was no evidence of an association between β-cell function (HOMA2-%B) and response (response reduction per SD higher 0.16 [95% CI −1.49, 1.81] mmol/mol; P = 0.85). Islet autoantibody prevalence was low (2.8% GAD or IA2 positive; response reduction for presence of autoantibodies: 5.6 [95% CI −3.6, 14.7] mmol/mol; P = 0.23).

In PRIBA, higher triglycerides was associated with reduced glycemic response (response reduction per SD increase 2.54 [95% CI 0.99, 4.08] mmol/mol; P < 0.001), with a consistent direction of association for higher BMI (response reduction per higher BMI 0.96 [95% CI −0.54, 2.46] mmol/mol; P = 0.21) and lower SHBG (response reduction per SD higher SHBG −1.19 [95% CI −2.81, 0.42] mmol/mol; P = 0.15) (Fig. 1 and Supplementary Table 1). In CPRD, higher triglycerides and BMI were associated with reduced HbA_1c response (Fig. 1 and Supplementary Table 1). HDL-c was not associated with response in either data set (P = 0.81 in PRIBA; P = 0.46 in CPRD).

Results were consistent when 1) stratifying by sex (Supplementary Table 1), 2) controlling for baseline HbA_1c, age at therapy, sex, duration of diabetes, eGFR, LDL-c, ethnicity (CPRD only), and cotherapy change (PRIBA only) in multivariable analysis of each data set (Supplementary Table 2), 3) in PRIBA controlling for fasting glucose (Supplementary Table 3), and 4) in CPRD with 12-month HbA_1c response as the outcome (Supplementary Table 4).

Higher triglycerides were associated with reduced glycemic response independently of BMI in both data sets, and higher BMI was associated with reduced response independently of triglycerides in CPRD (Supplementary Table 5). To examine the potential clinical implication of this finding, we compared mean baseline HbA_1c adjusted response in three patient subgroups defined by standard clinical cutoffs for obesity (BMI ≥30 kg/m²) and high triglycerides (≥2.3 mmol/L) (subgroup A: nonobese and normal triglycerides; subgroup B: nonobese or normal triglycerides; and subgroup C: obese and high triglycerides).

In PRIBA, we found mean 6-month baseline HbA_1c standardized glycemic response was halved for the obese and high triglycerides subgroup (subgroup C −5.2 [95% CI −1.8, −8.6] mmol/mol [−0.5% (95% CI −0.2, −0.8)]) compared with the nonobese and normal triglycerides subgroup (subgroup A −11.3 [95% CI −8.4, −14.1] mmol/mol [−1.0% (95% CI −0.8, −1.3)]) and was significantly reduced compared with intermediate subgroup B (−9.9 [95% CI −7.6, −12.2] mmol/mol [−0.9% (95% CI −0.7, −1.1)]) (Fig. 2A). Direction of effect was replicated in CPRD, albeit with smaller differences in mean response between subgroups (subgroup A mean baseline adjusted HbA_1c response −10.3 [95% CI −9.8, −10.7] mmol/mol [−0.9% (95% CI −0.9, −1.0)]; subgroup B −8.8 (95% CI −8.5, −9.1) mmol/mol [−0.8% (95% CI −0.8, −0.8)]; and subgroup C −7.5 (95% CI −7.0, −7.9) mmol/mol [−0.7% (95% CI −0.6, −0.7)]) (Fig. 2B).

A total of 15,616 patients were followed up in this analysis for a mean time of 1.5 years. Over the 3-year study period, 3,514 (23%) patients had glycemic failure (confirmed HbA_1c ≥69 mmol/mol [8.5%]). We observed an increased relative risk of glycemic failure (reflecting a less durable response) in the same obesity- and high triglycerides–defined subgroups, standardizing for other clinical characteristics (HRs for glycemic failure: subgroup C, obese and high triglycerides vs. subgroup A, nonobese and normal triglycerides, 1.28 [95% CI 1.16, 1.41], P < 0.001; subgroup B, obese or high triglycerides vs. subgroup A, 1.17 [95% CI 1.08, 1.27], P < 0.001; and subgroup C vs. subgroup B, 1.09 [95% CI 1.01, 1.18], P = 0.04) (Supplementary Table 6). Consistent relative differences between subgroups were observed at HbA_1c failure thresholds of 7.5% and the baseline HbA_1c specific to each individual patient (Supplementary Tables 7 and 8). These results translated into significant differences between subgroups in the absolute probability of glycemic failure at 3 years (subgroup C: obese and high triglycerides, 39% [95% CI 37, 42]; subgroup B: obese or high triglycerides, 37% [95% CI 35, 38]; and subgroup A: nonobese and normal triglycerides, 32% [95% CI 31, 34]) (Supplementary Fig. 2).

We found no evidence of an association between any marker of insulin resistance and 6-month glycemic response to GLP-1 receptor agonists in PRIBA (n = 339) or CPRD (n = 4,464) in continuous analysis (Fig. 3 and Supplementary Tables 8 and 9). There was also no evidence for a difference in response to GLP-1 receptor agonists across the obesity- and triglyceride-defined subgroups (all subgroup comparisons, P > 0.40) (Supplementary Table 10 and Supplementary Fig. 3), although there were few subjects in the nonobese, normal triglyceride subgroup starting GLP-1 receptor agonist therapy in both data sets (PRIBA, 2% and CPRD, 5%). Similarly, in CPRD, we found no evidence of an association between durability of glycemic response and BMI (HR per unit increase 1.01 [95% CI 1.00, 1.02]; P = 0.29) or triglyceride levels (HR per unit increase 0.99 [95% CI 0.95, 1.04]; P = 0.80) (Supplementary Table 11) or of a difference in durability of response across obesity- and triglyceride-defined subgroups (Supplementary Table 12 and Supplementary Fig. 4).

Our results show that markers of higher insulin resistance are consistently associated with reduced glycemic response to DPP-4 inhibitor therapy. In our U.K.-representative cohort, 22% of patients were obese with high triglycerides (≥2.3 mmol/L), and these patients had both markedly reduced short-term glycemic response and shorter durability of response on DPP-4 inhibitor treatment. With GLP-1 receptor agonists, we found no evidence of an association between markers of insulin response and either 6-month glycemic response or durability of response to 3 years. Findings were robustly demonstrated in a prospective study and validated in real-world data and provide a starting point for the application of a precision diabetes approach with DPP-4 inhibitor therapy.

Strengths of this study include that we have shown consistent findings across several clinical features and markers of insulin resistance in a prospective study and large data set of electronic health care records. We have shown that findings are robust with adjustment for baseline HbA_1c (13,24) and potential confounders and by definition of glycemic response, with similar associations for short-term (6- and 12-month) and long-term (3-year durability) glycemic outcomes. Our study is the first to identify characteristics associated with durability of response to DPP-4 inhibitor therapy, an area in which evidence is limited (25).

Limitations of this study include that we were only able to partially replicate our results from the PRIBA study cohort, as measures such as C-peptide were not available in our replication data set. Our effect size for triglycerides is notably smaller in our replication data set. It is possible this relates to differences in triglyceride measurement (we were unable to confirm if measured triglycerides were fasted in these real-world data), to increased error in electronic health care records in comparison with the prospective study (26), or to the effect of statistical chance in the smaller data set. The only long-term follow-up data we had to evaluate durability of glycemic response was from the routine primary care data set CPRD; further evaluation in a trial setting with greater follow-up than PRIBA would be of considerable interest. An additional important limitation is that this study has examined response to only two of the available therapies. Evidence is limited for other therapies, although a previous study found no evidence of a relationship between clinical insulin resistance or dyslipidemia markers and glycemic response with the sodium–glucose cotransporter 2 inhibitor dapagliflozin (27). High BMI and triglycerides have both been shown to be associated with modest increases in the rate of diabetes progression (28). Although this is unlikely to be relevant to our finding for 6-month glycemic response, this could influence our findings for treatment durability, and replication looking at other comparison therapies is therefore particularly important in this context. Although we have only examined relatively crude measures of insulin resistance, for clinical practice we consider it very unlikely that more complex measures would ever be feasible (29).

Existing studies of the association between insulin resistance and short-term glycemic response to DPP-4 inhibitors have not shown consistent findings and are constrained by methodological and reporting limitations, as recently reviewed by Bihan et al. (13). Meta-regression of study-level data has suggested reduced glycemic response in patients with higher BMI in one study (30), but no relationship in another analysis (31). These studies should be interpreted with some caution due to risk of ecological bias (32,33). A number of individual clinical trials of DPP-4 inhibitors have commented on consistency of glucose response across subgroups defined by baseline BMI or insulin resistance; reduced glycemic response with high HOMA2-IR was reported in two of seven studies and reduced response with high BMI in 6 of 36 studies, as reviewed in Bihan et al. (13). No studies reported an opposite direction of effect. These reports are very limited, with the vast majority providing no statistical comparison or details of what analysis was undertaken. An important issue for analysis of this nature is accounting for the influence of baseline HbA_1c, the strongest predictor of glycemic response, which may confound true associations, especially as baseline HbA_1c and insulin resistance are positively correlated (24,34,35). There are limited data examining the relationship between triglycerides and response to DPP-4 inhibitors; however, one study stratified patients by baseline triglycerides (</>1.7 mmol/L) and found the odds of achieving an HbA_1c target of 53 mmol/mol (7%) were doubled in the low-triglyceride subgroup (odds ratio 2.2 [95% CI 1.0, 4.7]; P = 0.04) (36).

Although it is plausible that our finding of reduced glycemic response in those with high BMI or high triglycerides directly relates to insulin resistance through reduced effect of drug-potentiated insulin secretion, this effect is not apparent in other drugs with effects on insulin secretion; for example, there is no relationship between obesity and response to sulfonylurea therapy or GLP-1 receptor agonists (15,37). An alternative explanation would be a direct effect of lipotoxicity, or indirect associations with other (unmeasured) factors important to DPP-4 inhibitor response. A direct mechanism for lipotoxicity in reducing response to incretin-based therapy has been previously suggested, with expression of GLP-1 receptors diminished in islets exposed to elevated fatty acid levels in animal models and β-cell response to GLP-1 restored following fatty acid reduction with fibrate pharmacotherapy; however, this mechanism would not explain the lack of an association between these features and GLP-1 receptor agonist response (38). It has also been shown that GLP-1 response is blunted in obese insulin-resistant patients with high liver fat and also blunted in patients with high fasting triglycerides (39,40); therefore, impaired GLP-1 secretion in obese insulin-resistant individuals represents a potential indirect mechanism that could also account for the lack of a similar relationship for injected GLP-1 receptor agonist therapy. Although the lack of association for HDL-c may be considered unexpected, we note HDL-c has a much weaker relationship with insulin resistance than either triglycerides or fasting insulin/C-peptide, which may explain this finding (18).

Our findings have potential implications for clinical practice, as both BMI and triglycerides are routinely available at no additional cost. Stratification of treatment based on these criteria may therefore be cost effective even with the more modest differences in treatment effect seen in our replication cohort. Although our own and previous research suggests these findings may be specific to DPP-4 inhibitors, further work examining the relationship between these and other factors and response to comparator drugs is needed. Our study design, emphasizing the importance of replication across data sets, provides an exemplar for such future analyses. In addition, although simple categorization by subgroup may provide a starting point for prediction of therapy response in type 2 diabetes, we anticipate a more sophisticated precision diabetes approach combining features into a multivariable response calculator will have greatest clinical utility, and this is an important area for future research (8,41).

In conclusion, our study shows simple markers of higher insulin resistance are consistently associated with reduced glycemic response to DPP-4 inhibitor therapy. This finding was robustly demonstrated in a prospective study, was validated in real-world data, and provides a starting point for the application of a precision diabetes approach to DPP-4 inhibitor therapy in type 2 diabetes.

Acknowledgments. The authors thank the staff of the NIHR Exeter Clinical Research Facility and NIHR Diabetes Research Network for assistance with conducting the study. The authors also thank the members of the PRIBA study group (see Supplementary Data for participating centers and members) and all cohort participants.

Funding. The PRIBA study was funded by a National Institute for Health Research (U.K.) Doctoral Research Fellowship (DRF-2010-03-72 to A.G.J.) and supported by the NIHR Clinical Research Network. The MASTERMIND Consortium is supported by the Medical Research Council (U.K.) (MR/N00633X/1). B.M.S., A.V.H., B.A.K., and A.T.H. are supported by the NIHR Exeter Clinical Research Facility. T.J.M. is an NIHR Chief Scientist Office Clinical Senior Lecturer. N.S. acknowledges support by Innovative Medicines Initiative Joint Undertaking under grant agreement 115372, the resources of which comprise financial contributions from the European Union’s Seventh Framework Programme (FP7/2007–2013) and EFPIA (European Federation of Pharmaceutical Industries and Associations) companies’ in-kind contribution. R.R.H. and A.T.H. are NIHR Senior Investigators. E.R.P. is a Wellcome Trust New Investigator (102820/Z/13/Z). A.T.H. is a Wellcome Trust Senior Investigator. A.G.J. is supported by an NIHR Clinician Scientist award.

The views expressed are those of the authors and not necessarily those of the National Health Service, NIHR, or Department of Health. The funders had no role in any part of the study or in any decision about publication.

Duality of Interest. W.E.H. declares a grant from Quintiles. E.R.P. declares personal fees from Eli Lilly and Company, Novo Nordisk, and AstraZeneca. N.S. declares personal fees from Boehringer Ingelheim, Eli Lilly and Company, Novo Nordisk, and Janssen and a grant from AstraZeneca. R.R.H. declares research funding from Bayer, AstraZeneca, and Merck Sharp & Dohme and honoraria from Amgen, Bayer, Elcelyx, Janssen, Intarcia Therapeutics, Merck Sharp & Dohme, Novartis, Novo Nordisk, and Servier. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.M.D. and A.G.J. designed the study and drafted the article. B.M.S., L.R.R., and M.N.W. extracted the CPRD data. A.V.H., B.A.K., T.J.M., and A.G.J. researched the data (PRIBA study). J.M.D., B.M.S., and A.G.J. analyzed the data with assistance from W.E.H. W.E.H., N.S., R.R.H., E.R.P., and A.T.H. discussed and contributed to study design and provided support for the analysis and interpretation of results. All authors critically revised the article and approved the final version. A.G.J. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the Diabetes UK Professional Conference, Manchester, U.K., 8–10 March 2017.