Background Offering genetic testing for Maturity Onset Diabetes of the Young (MODY) to all young patients with type 2 diabetes has been shown to be not cost-effective. This study tests whether a novel algorithm-driven genetic testing strategy for MODY is incrementally cost-effective relative to the setting of no testing.
Methods A decision tree was constructed to estimate the costs and effectiveness of the algorithm-driven MODY testing strategy and a strategy of no genetic testing over a 30-year time horizon from a payer’s perspective. The algorithm uses glutamic acid decarboxylase (GAD) antibody testing (negative antibodies), age of onset of diabetes (<45 years) and body mass index (<25 kg/m2 if diagnosed >30 years) to stratify the population of patients with diabetes into three subgroups, and testing for MODY only among the subgroup most likely to have the mutation. Singapore-specific costs and prevalence of MODY obtained from local studies and utility values sourced from the literature are used to populate the model.
Results The algorithm-driven MODY testing strategy has an incremental cost-effectiveness ratio of US$93 663 per quality-adjusted life year relative to the no testing strategy. If the price of genetic testing falls from US$1050 to US$530 (a 50% decrease), it will become cost-effective.
Conclusion Our proposed algorithm-driven testing strategy for MODY is not yet cost-effective based on established benchmarks. However, as genetic testing prices continue to fall, this strategy is likely to become cost-effective in the near future.
- algorithm-driven genetic testing
- maturity onset diabetes of the young
- gene panel testing
Statistics from Altmetric.com
- algorithm-driven genetic testing
- maturity onset diabetes of the young
- gene panel testing
Maturity Onset Diabetes of the Young (MODY) is a monogenic form of diabetes that accounts for up to 2% of all diabetes cases.1 It is characterised by the onset of diabetes under the age of 25 years, autosomal dominant inheritance and non-insulin dependence.2 To date, mutations in up to 14 genes have been identified to cause MODY, of which the majority of the MODY cases have been attributed to one of the following four genes: HNF1A, HNF4A, GCK and HNF1B. Each mutation, depending on its downstream effect on the insulin pathway, may provide insight into modifying diabetes treatment. For instance, HNF1A and HNF4A mutations respond better to sulphonylureas than other treatments including metformin (usual first-line medication for type 2 diabetes (T2D)).3 Patients who are previously diagnosed with insulin-requiring diabetes and require regular monitoring of their glycaemic control may be successfully switched to oral sulphonylurea tablets on diagnosis of HNF1A/HNF4A MODY, obviating the need for regular self-monitoring of blood glucose.4 Adapting treatment plans to include sulphonylureas may improve quality of life by reducing the need for daily injections. It is also likely to save costs given that sulphonylureas are less expensive than insulin.
Recent advancements in next-generation sequencing (NGS) technology afford the opportunity to screen multiple MODY genes at one time for an estimated cost of US$1000. However, prior to implementing this new clinical tool, it is necessary to understand whether the benefits of genetic testing and targeted treatment are likely to justify the cost. In a recent US-based study, simultaneously testing for the three most common MODY genes (HNF1A, HNF4A and GCK) among all newly diagnosed patients with T2D aged 25–40 years was shown not to be cost-effective with a cost-effectiveness ratio of US$205 000 per quality-adjusted life year gained.5 However, the authors suggested that such universal testing would be cost-effective if MODY prevalence increased from 2% to 6% in the target population. One strategy for increasing the prevalence rate of the target population is to identify a subset of patients most at risk for MODY prior to gene testing.
This study aims to customise and extend the prior analysis to assess whether a cost-effective algorithm-driven genetic testing strategy for MODY could be identified by better targeting the subset of the population identified for genetic testing. The algorithm involves testing for beta-cell autoantibodies, and then among those who test negative, identifying a subset who are less than age of 30 years or who are between the ages of 30 and 45 years with body mass index (BMI) of less than 25 kg/m2. MODY genetic testing is done only for this subset. Using cost and prevalence data from Singapore, we test whether this new algorithm-driven genetic testing approach is cost-effective relative to no testing.
Traditional criteria for gene screening include a family history of diabetes and diabetes onset under age of 25 years; however, using these criteria alone, up to 50% of MODY patients will not have fulfilled the criteria for screening.6 To be more inclusive, our proposed strategy starts with considering individuals diagnosed with diabetes at age of 45 years or lower.
Excluding type 1 diabetes (T1D)
The algorithm first identified people with T1D in the starting population. Those who tested positive for GAD antibodies Ab (GADAb) were classified as having T1D or latent autoimmune diabetes of adulthood. GADAb was analysed in the core laboratory (IASP 2015 Lab number 1501), Lee Kong Chian School of Medicine, by radioligand assay using 35S-labelled full-length GAD65 according to a harmonised protocol.7 GADAb positivity was determined as the 99.5th percentile of 1192 healthy German control subjects (age range 18–70 years; mean age 39.7 years) and 145 healthy Singaporean control subjects (age range 20–69 years; mean age 49.1 years). Greater than or equal to 21 DK units/mL for GADAb was defined as positive. Those who tested positive for GADAb did not undergo further genetic testing and would continue with their current treatment. While conceding that MODY has been found in individuals previously misdiagnosed as having T1D,8 and beta cell autoantibody positivity has been found in those with MODY, this was an uncommon occurrence9 (<1%) and associated with GADAb positivity rather than IA-2A Ab.9 In Asians, previous studies have consistently demonstrated the low prevalence of GADAb positivity in those with clinically classified T1D compared with Caucasian populations (40% versus >84%10–13). GADAb appears to be the most prevalent antibody among Asian autoantibody-positive T1D patients, even in studies incorporating other antibodies like the zinc transporter 8 antibody.14 We anticipate that in our Asian cohort, the low background incidence of T1D (2.46 per 100 000 in Asian cohorts compared with 60 per 100 00015 in Caucasian populations16) and a low prevalence of beta cell autoimmunity among those with clinically diagnosed T1D point towards a greater likelihood of T1D rather than MODY in an Asian patient with diabetes and positive GADAb.
Subgroup targeted for MODY testing
Patients who are GADAb negative were classified as non-T1D. While this group will inevitably include those with GADAb-negative T1D, we felt that it was important to include these individuals for genetic testing given the heterogeneity of diabetes aetiologies and overlapping clinical phenotypes in Asians.17 Among these patients, genetic testing for MODY was targeted at those with features that are less consistent with T2D. These included those with diabetes onset under age of 30 years (ie, very young onset of T2D) or patients with diabetes diagnosed at age between 30 and 45 years and BMI less than 25kg/m2 (ie, young onset of T2D but slim).
We used a NGS gene panel that contains known/putative genes associated with MODY (ie, HNF1A, HNF4A, GCK, HNF1B, INS, KNCJ11, ABCC8, IPF1, CEL, NEUROD1, PAX4, KLF11, BLK, INSR, LMNA and PPARG). The 14th MODY gene APPL1 was not included in this panel as it was discovered only after this panel and study was designed.18 Those who tested negative for all 16 genes were classified as having T2D and continued their pre-existing medications in the same ratio as per usual treatment. Those who tested positive for a particular gene received tailored treatment (or no treatment) depending on which gene was identified. In particular, for those who tested positive for HNF1A and HNF4A, sulphonylurea (oral) treatment was given, while those who tested positive for GCK did not receive medication. For the rest of the rarer MODY genes, treatment was tailored according to mutation (eg, patients with INS MODY required insulin treatment, as did patients with HNF1B MODY). Patients aged between 30 and 45 years who had BMI greater than 25 continued with their pre-existing treatment.
This study was approved by Singhealth Centralized Institutional Board (No. 2013/651/C).
Decision tree model
A decision tree was constructed to estimate the incremental costs and effectiveness of the algorithm-driven testing strategy relative to no genetic testing over a 30-year time horizon (figure 1). We chose a decision tree model due to the non-recursive nature of genetic testing for MODY as well as the rigour (yet simplicity) of this model in capturing the costs and effectiveness associated with the strategies considered. No genetic testing was chosen as the comparator since genetic testing for MODY is not currently performed as standard of care in Singapore. Patients in the no testing arm are treated with one of the usual treatment options. These involve insulin injections, oral glucose-lowering agents or controlling blood glucose levels by diet and exercise. The algorithm-driven testing strategy models the testing and treatment pathways described above. Our model does not consider future diabetes-related complications associated with MODY or T2D because inclusion of complications in the model would not alter our incremental cost-effectiveness ratio (ICER) estimate. The algorithm-driven genetic testing does not influence the risks of developing diabetes-related complications and patients’ survival, and thus, as our cost-effectiveness analysis compares the algorithm-driven genetic testing versus no testing for the same study population, effects of complications, if any, would be cancelled out. In fact, Naylor et al 5 included diabetes-related complications in their model and found that genetic testing had no effect on the risks of complications and life expectancy.
Prevalence parameters displayed in table 1 were estimated based on a sample of 4630 patients who were diagnosed with diabetes under age of 45 years and enrolled in one of two diabetes cohorts in Singapore: the Singapore Diabetes Cohort Study (SDCS) and the Singapore Young Adults with Diabetes (SYD) Study. SDCS consists of patients aged 21–75 years old with diabetes who were recruited from four public healthcare hospitals and six public sector primary care clinics in Singapore. SYD consists of patients recruited from a single hospital who were diagnosed with diabetes under age of 45 years.
The prevalence of positive GADAb in this sample was 6.8%. The negative GADAb group (93.2% of the sample) consists of three subgroups: (1) patients with diabetes onset under age of 30 years (11.8%); (2) patients with diabetes diagnosed between age of 30–45 years and BMI less than 25 kg/m2 (29.7%); and (3) patients who were diagnosed between age 30–45 years and have a BMI greater than 25 kg/m2 (51.7%). According to the algorithm, patients from subgroups (1) and (2) are most likely to have MODY mutations. Among these patients, 687 (36%) with sufficient stored serum and DNA provided samples for DNA analysis. These samples were sequenced using a target-enriched exon-capture NGS 16-gene panel. The test results indicated that overall MODY prevalence rate was 3.93% (2.3% for HNF1A/HNF4A, 0.9% for HNF1B, 0.43% for GCK and 0.3% for the 12 rare genes). There were no significant differences in BMI and age of onset of diabetes between those who were tested versus those who were not tested (BMI: 24.4+4.3 vs 24.0+4.3 kg/m2; age of onset of diabetes: 34.2+7.3 vs 35.0+8.2 years, both p>0.05). All variants found on NGS were confirmed with Sanger sequencing. We assigned pathogenicity only to variants that have been previously reported as implicated in monogenic diabetes and that are usually considered to be clearly pathogenic as per the Association for Clinical Genomic Science’s (ACGS) best practice guidelines.19 Those with unknown pathogenicity have been categorised as variants of unknown significance until further published reports from continued sequencing efforts allows a reclassification of their status. For those who continue to receive usual care, the proportion of patients who adopt each treatment strategy (ie, insulin treatment, oral glucose lowering agent and diet control) are based on best estimates for Singapore based on provider experience.
We assume that screening based on demographic variables (age and BMI) does not involve any additional costs as capturing these variables is part of standard practice. GADAb screening is assumed to occur in the first year of the study period and costs US$59.20 Annual treatment costs are the cost of average daily dosage of medicine/insulin injection based on restructured hospital pricing to patients20 multiplied by 365 days. The discounted sum of annual costs over a 30-year period is used to generate the total costs of treatment. The use of glucose strips to measure blood sugar levels is assumed to occur twice daily over the 30-year horizon among insulin-dependent patients. All costs are presented in US dollars (using exchange rate as on 18 November 2015: S$1=US$0.7) and discounted at 3.5% per year.21 It is assumed that those who do not require any treatment, including those who control their diabetes with diet and exercise only, incur no additional costs.
Effectiveness is measured in terms of quality-adjusted life years (QALYs), a measure of a person’s length of life weighted by utility, a valuation of their health-related quality of life. A utility value of 1 indicates a year spent in perfect health. In our analysis, patients who have T2D are assigned a utility value of 0.75 if they are on insulin treatment, 0.82 if on oral medications and 0.82 if on diet and exercise only, as obtained from Zhang et al. 22 A utility value of 0.92, which was estimated for a non-insulin diabetes patient without diabetes complications and without risk factors for cardiovascular diseases in Zhang et al 22 is used as a proxy for diabetes patients with no treatment. All utility values are discounted at 3.5% per year.21
We use the decision tree to estimate the costs and effectiveness of the two strategies. The ICER is the difference between the overall costs of the two strategies divided by the difference in QALYs. To account for uncertainty, one-way and probabilistic sensitivity analyses are conducted. These analyses model the impact of a change in any of the costs or probabilities on the ICER. In the one-way sensitivity analysis, all relevant costs and probabilities are varied over a reasonably large range, namely, ±25%.23 We also conduct threshold analysis to identify the threshold values of prevalence and costs that render the algorithm-driven MODY testing strategy cost-effective. In probabilistic sensitivity analysis, we assign distributions to all input parameters and perform 10 000 Monte Carlo simulations where each simulation generates an ICER of the algorithm-driven genetic testing relative to no testing. We assume beta distributions for all probabilities, normal distributions for utilities and gamma distributions for testing and treatment costs.23 Results are presented graphically showing the probability that genetic testing is cost-effective for a range of cost-effectiveness thresholds that decision makers may consider. Furthermore, to establish the robustness of our results, we vary the utility values used in our analysis by using those reported in other studies.
Base case cost-effectiveness analysis
The base case cost-effectiveness results are presented in table 2. The cost of the genetic test is US$1050. Total cost of the algorithm-driven testing strategy is, on net, higher than that of the no testing strategy by US$462 (US$9888 vs US$9426). The algorithm-driven testing strategy generates slightly more QALYs on average: 15.215 versus 15.210. These additional QALYs result from the reduced need for insulin treatment in the subset of patients with GCK mutation and the switch from insulin to oral sulphonylurea for patients with HNF1A or HNF4A mutations. Combining the cost and effectiveness estimates, the ICER for the algorithm-driven genetic testing strategy relative to no testing is US$93 663/QALY.
Results of the one-way sensitivity analyses (figure 2) show that the ICER is most sensitive to the prevalence of MODY mutations (ie, HNF1A/HNF4A, GCK and HNF1B) in the target population (ie, those who receive the gene panel), currently estimated to be 3.93% for Singaporean patients whom we identify should get the algorithm. It is also sensitive to the proportion of patients receiving insulin treatment under usual treatment. The cost-effectiveness acceptability curve (figure 3) shows that, for a commonly applied willingness-to-pay threshold of US$50 000/QALY, the algorithm-driven testing strategy is currently cost-effective in 30% of iterations. The threshold analysis further reveals that provided all other factors remain constant, if the prevalence of HNF1A/HNF4A mutation among those undergoing the gene panel test increases to 4.1% (currently 2.3%) or the prevalence of GCK mutation increases to 0.85% (currently 0.43%), or the cost of the genetic test drops below US$530, then the algorithm-driven MODY testing strategy would be cost-effective using the threshold of US$50 000/QALY.
As Zhang et al’s study22 is based on a sample of patients with diabetes aged 62 years on average while our study population consist of younger people (aged 45 years and below), we conduct two additional sensitivity analyses to examine the robustness of our results with respect to our use of utility values from Zhang et al.22 In the first test, we use utility values by treatment type from another study, namely, Redekop et al. 24 As that study does not provide a utility value for diabetes with no treatment, we assign a utility of 0.84 for no treatment (as reported in Ara and Brazier25). The value of ICER increases only slightly in the sensitivity analysis (ie, US$93 663/QALY to US$99 469/QALY). In the second test, we attempt to estimate the utilities by treatment type for the age group less than 45 years using data from Zhang et al 22 and Redekop et al. 24 To do this, we first use the utility value for age group 45 years and below reported in Zhang et al 22 and assume that to be the utility for oral therapy. (Note that Zhang et al 22 reported utility by age group but not by age-type of treatment needed for our model.) We then assign utility values for insulin, diet and no treatment such that their differences from the utility value for oral therapy match that reported in Redekop et al. 24 Specifically, difference is 0.13 between oral therapy and insulin, 0.03 between oral therapy and diet and 0.08 between oral therapy and no treatment. We rerun the model using these utility values. The results are similar to our base case estimate.
This study reported the results of a cost-effectiveness analysis comparing an algorithm-driven testing strategy to detect MODY versus no testing using Singapore as a case study. We found that compared with the no testing strategy, the algorithm-driven MODY test incurred higher costs while generating slightly higher QALYs, resulting in an ICER of US$93 663 per QALY. The cost difference (ie, US$462) results from the GADAb screening costs (US$59) and MODY testing cost (US$1050) offset by savings from (1) patients with MODY mutations switching to cheaper oral medications or no treatment (from more expensive insulin treatment) and (2) the elimination of the need for glucose strips for self-blood glucose monitoring. The greater utility is due to the switch to oral medications/no treatment. This ICER of US$93 663 per QALY is not cost-effective using the conventional threshold of US$50 000/QALY, but our analysis indicates that if the cost of MODY testing were to decrease to US$530, the algorithm-driven MODY testing would become cost-effective. This is likely to happen in the near future as costs of genetic tests continue to fall.26
Our study differs from Naylor et al,5 who found an ICER of US$205 000 per QALY, in two important ways. First, the testing strategy considered in Naylor et al 5 is not algorithm based; all individuals with T2D aged between 25 and 40 years are subjected to MODY testing. Using the algorithm-based approach, we were able to identify a subgroup of patients most likely to test positive for MODY, thus reducing screening and testing costs and increasing the chances of finding MODY cases. This approach generates significant cost savings relative to the strategy of universal MODY testing.
The second difference between our study and Naylor et al 5 concerns the testing and treatment costs. In Naylor et al,5 the cost of conventional Sanger single gene testing for the three most common MODY subtypes (HNF1A, GCK and HNF4A) was reported to be US$2580 per patient. This cost in Singapore is US$1050 due to the use of NGS technology, which allows screening of multiple genes (16 in this panel) simultaneously. Their treatment costs are also higher than those used in our study. Cost of insulin treatment (including self-blood glucose monitoring) is almost twice the cost in Singapore (US$2641 vs US$1528). Costs for oral medications are nearly 12 times greater (US$767 vs US$61), and costs of sulphonylureas are 1.5 times greater (US$96 vs US$61). To estimate the influence of these cost differences on the ICER, we re-estimated our model using the US testing and treatment cost data as well as utility values used in their study. The resulting ICER is US$148 050 per QALY. This ICER is higher than the ICER estimated for Singapore but is still substantially lower than the ICER for the US estimated in Naylor et al. 5 This suggests that the relatively low ICER obtained in our study and the relatively high ICER in Naylor et al 5 are partly due to the higher cost of genetic testing and treatment in the USA compared with Singapore. It also confirms the value of our proposed algorithm in reducing the testing costs, leading to more favourable cost-effectiveness estimates. In fact, using the US costs and utilising our proposed algorithm, the cost of genetic testing would only need to be reduced from US$2580 to US$1066 for the genetic test to become cost-effective in the USA. Without the algorithm, it would have to be reduced further to US$700 for the genetic test to be cost-effective.5
Our study has a number of limitations. First, our algorithm has not been validated in other population groups or ethnicities or in other cohorts in Singapore. Asians tend to have lower BMI values, different distributions of fat mass and different rates of diabetes for any given BMI as compared with Caucasians.27 28 As such, results would not be expected to generalise to Caucasians. Moreover, the prevalence of the MODY gene appears to vary by ethnicity even among Asians. Further analyses on additional cohorts within and beyond Singapore are needed to confirm these results and to test whether the algorithm-driven testing strategy would be cost-effective in other settings. Our model also did not include possible improvement in glycaemic control associated with optimal treatment following a diagnosis of MODY. Although possible, efficacy of sulfonylurea on glycaemic control is mixed. Shepherd et al 4 and Thanabalasingham et al 6 reported an HbA1c fall of 0.8%–1.5% in small numbers of patients (n=6–8 per study) following a switch to sulfonylurea therapy, but follow-up was limited to an average of 3 months in one study6 and a median duration of 6 months in the other.4 Bacon et al 29 did not find a significant fall in HbA1c when switching patients with HNF1A MODY from insulin to sulfonylureas. Therefore, we did not model better glycaemic control as an outcome of MODY testing. We used utility data based on a group of older patients than those in our study population (62 vs 45) due to the lack of utility data for younger diabetes patients by treatment type in the literature. However, our sensitivity analyses show that our base cost-effectiveness estimates change little when we used utility values from other studies as well as when we used our estimated utility values for lower age groups. Finally, overall MODY prevalence in our sample population was estimated to be only 0.58% (27/4630), which is lower than in previous studies.30 If the true prevalence in this or other populations is greater, then MODY testing becomes more cost-effective.
In conclusion, in combination with a simple algorithm to select individuals for targeted screening, NGS panel sequencing could soon be a cost-effective strategy to tailor diabetes treatment based on genetic variants.
Contributors HVN led the model development and contributed to data analysis; SM contributed to data analysis; EF and DS-LG conceptualised the initial idea; DS-LG conducted clinical investigations that provided model inputs. All authors contributed to manuscript writing and manuscript revisions. HN and DS-LG are the overall guarantors of the manuscript.
Funding DS-LG and EF received funding from the Stratified Medicine Programme Office, a joint initiative by the Biomedical Research Council and the National Medical Research Council, Singapore.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.