To estimate the cost-utility of the targeted PRS-based strategy when compared to the usual strategy in Finnish adult population, an individual-level state-transition model was constructed. The developed model included four mutually exclusive health states (i.e., Healthy, T2D, T2D with complications, and Death). Possible state-transitions between the modelled health states are illustrated in Figure 1. In the model, individuals developing T2D could develop further T2D-related complications, they might die, or they might survive to the next year in the T2D state (i.e., 1-year cycle length was applied in the model) without any event. In the present study, the individual-level modelling approach was applied to overcome the Markov assumption (Briggs et al., 2006), which relates to the fact that future transitions between health states are not dependent on previous states. Therefore, the individual-level state-transition modelling was applied to incorporate the memory of events occurring for simulated individuals in the model. Finally, the developed model was used to estimate the expected number of new T2D cases and associated consequences (in terms of costs and QALYs) with and without using PRS-based information to improve targeting of lifestyle and medical intervention using a lifetime time horizon.

In the developed model, individuals were passed through the model one-by-one, their results were stored and then the expected values of a cohort were obtained by aggregating the individual results. To reflect adequately both stochastic variability (i.e., individual-level variation) and parameter uncertainty (Degeling et al., 2017), in total 1,000 independent cohorts with 20,000 individuals each (i.e., 10,000 individuals per study arm) were simulated to obtain estimates for a cost-utility analysis. The model was written in R (R Core Team, 2021). See Supplementary Material S1 for further technical details.

In the base-case scenario, the model was populated based on the real-world characteristics of the Finnish population aged 30–79 years without T2D at baseline (n = 2.97 million Finnish adults in 2017) but having high overall risk (i.e., 10%–20%) for developing T2D during the next 10 years measured with the FINDRISC. Based on these criteria, a total of 10.5% (n = 313,000) of the Finnish adult population was estimated to belong to the target population of the present study. At baseline, the average age of this target population was 61.5 years and 63.9% were men. The baseline distribution of FINDRISC in the modeled target population is shown in Supplementary Material S2. In the present study, the FINDRISC score was divided in five age- and sex-specific categories (i.e., from low risk to very high risk) indicating the 10-year risk of T2D.

The baseline risk of T2D was estimated based on dataset obtained from The National FINRISK Study (n = 15,868), which was further enriched with a 10-year data from the national medicine reimbursement registry maintained by the Social Insurance Institution of Finland. From this register-based dataset, all new reimbursement rights and/or the first purchases for T2D medicines were recorded and used as an indicator for a T2D diagnosis in the final analysis dataset. Finally, a parametric survival regression modelling was applied to estimate the 10-year risk of T2D based on baseline age, sex, and FINDRISC, which in turn was estimated based on age, body mass index (kg/m2), use of blood pressure medication, history of high blood glucose, physical activity, daily consumption of vegetables, fruits, or berries, as well as family history of diabetes (Lindström and Tuomilehto, 2003). Among the applied parametric regression models, the Weibull survival regression model was considered to provide the most reliable fit according to the AIC and BIC criteria, as well as visual inspections of the estimated survival curves. In addition, similar statistical analyses were also conducted using PRS as an additional predictor (i.e., together with age, sex, and FINDRISC) to be able to assess the baseline risk of T2D with and without genetic risk information. In these analyses we used a previously published PRS, calculated with LDPred algorithm and validated and tested in the UK Biobank population (120,280 and 288,978 participants, respectively) (Khera et al., 2018). In the FINRISK dataset, the PRS was then summarized from 6.9 million genetic variants. For the statistical analyses, the PRS was standardized to a normal distribution with mean of 0 and standard deviation of 1. The PRS approach is described in more detail in a recently published study (Marjonen et al., 2021).

The use of parametric survival models enabled the extrapolation of event risks over the actual 10-year follow-up period and thus, annual age- and sex-specific transition probabilities applied in the developed individual-level state-transition model were estimated based on these estimated and extrapolated incidence rates. The coefficients of the Weibull regressions (with and without PRS) for the incidence of T2D are shown in Supplementary Material S3.

To model the risk of T2D-related micro- and macrovascular complications in persons with newly diagnosed T2D, a previously developed Weibull survival regression model was applied (Martikainen et al., 2021). The coefficients of the Weibull regression for the incidence of T2D-related complications is shown in Supplementary Material S4.

To model the risk of overall mortality, the national all-cause life tables for men and women were used to characterize the risk of death conditional on age and sex. In addition, the risk of death in the modeled “T2D” and “T2D with complications” health states were adjusted to consider the increased risk of death in those health states by applying previously published HRs (Taylor et al., 2013; Weir et al., 2016).

In the present study, individuals with a screened ≥ 20% 10-year overall risk for T2D (i.e., very high risk of T2D) were assumed to have been directed to participate a lifestyle intervention and an additional medical therapy, if eligible with the national reimbursement criteria for medicines used along with diet and exercise to help manage weight in adults.

The hypothetical lifestyle intervention was assumed to consist of diet changes and increase in physical activity leading to a modest (i.e., from 2.5% to 4.9%) weight loss over a period of 1 year (Dunkley et al., 2014; Cameron Sepah et al., 2017; Castro Sweet et al., 2018) and its effect was assumed to last for 15 years (Lindström et al., 2013). Weight loss has been shown to be significantly associated with a reduced risk of T2D (Li et al., 2008; Lindström et al., 2013). Therefore, the effectiveness of the lifestyle intervention used in the model was conveyed through weight loss in kg.

In Finland, a glucagon-like peptide-1 receptor agonist, liraglutide 3 mg daily in combination with diet and exercise is currently reimbursed for individuals with BMI ≥ 35 kg/m² and impaired glucose tolerance (IGT), as well as with medication treated hypertension or dyslipidemia. In addition, a naltrexone–bupropion combination therapy is also currently reimbursed for the treatment of obesity for individuals with BMI ≥ 40 kg/m² or BMI ≥ 35 kg/m² and with medication for T2D, dyslipidemia, or hypertension. Therefore, for simplicity, it was assumed that 30% of persons receiving a lifestyle intervention would also be applicable to receive a medical intervention which was estimated to result in a weight loss of over 5% (Lundgren et al., 2021). This proportion of individuals receiving a medical intervention was estimated based on The FinHealth 2017 Study results, where over 30% of persons aged over 55 and FINDRISC over 15 (i.e., high risk of T2D) had hyperglycemia (Koponen et al., 2018). Finally, the association between the weight loss and the reduction in the risk of T2D was estimated based on a post-hoc analysis of Finnish DPS study data which is described in detail in a previous study (Jalkanen et al., 2021). The estimated average risk reduction applied in the model was 26% (HR = 0.74). The clinical parameters used in the model are shown in Table 1.

In the present study, the societal perspective (excluding direct non-medical costs, such as travel costs associated with the utilization of health care services) was applied in the study. The excess primary health care cost estimates of T2D were obtained from a recent Finnish study (Martikainen et al., 2021). The excess secondary health care costs of T2D, its complications and T2D-related productivity losses for persons under the age of 65 were based on previous Finnish studies (Koski et al., 2017; Koski et al., 2018a; Koski et al., 2018b). In addition, the annual average (per-person) costs of liraglutide treatment (used as a proxy for all medical therapies) and T2D medication (ATC-code A10) costs were obtained from the national medicine statistics maintained by the Social Insurance Institution of Finland. The intervention costs of the lifestyle counselling intervention was based on a Finnish lifestyle coaching program for obesity (Väätäinen et al., 2019). Since the PRS testing is not generally available in the Finnish public health care, in the base case analysis it was assumed that the unit cost of a PRS test is 50 Euro per test. Finally, all costs were adjusted to the 2021 price level using the official health care price index determined by Statistics Finland. All applied cost estimates are presented in Table 2.

Utility estimates applied to estimate the number of quality-adjusted life years (QALYs) achieved in the target population were obtained from previously published studies (Table 3). Briefly, the published population-level EQ-5D-3L utility values (stratified by age and sex) were applied to represent the average health-related quality of life in the target population (Saarni et al., 2006; Martikainen et al., 2011). Furthermore, EQ-5D-3L-based disutility weights (Clarke et al., 2002; Bagust and Beale, 2005; Solli et al., 2010; Kontodimopoulos and Pappa, 2012; Beaudet et al., 2014) were applied to adjust the impacts of T2D and its complications on the health-related quality of life. These disutility values were estimated as a weighted average, where disutility values associated with a single complication were weighted by their observed incidences between years 2000 and 2017 in Finland (Arffman et al., 2020).

In the present study, QALY was considered as a primary effectiveness outcome. Expected average costs and QALYs were estimated for both interventions. In the cost-utility analysis, it was first checked if the targeted PRS-based strategy is a dominant option (i.e., having higher QALYs at a lower cost vs. the usual strategy). In case of non-dominance, an incremental cost-effectiveness ratio (ICER) was estimated.

In the base case, a life-time horizon, and a 3% discount rate per year for costs and QALYs (from the second year onward) were applied in accordance with the national HTA guidelines (Pharmaceuticals Pricing Board, 2019).

To study the robustness of the obtained results, different one-way and scenario sensitivity analyses were conducted to assess the impact of changes in the input parameters on outcomes:

1) Effect of applied discount rate on ICER was studied by varying discount rates between 0% and 5%

In addition, to test the association between the assumed price of the PRS testing and the ICER, a previous calculation framework (Standaert et al., 2014) was applied to indicate at what price the ICER equals zero (i.e., no difference in total cost between interventions). This point was referred as the cost neutral point. Finally, as mentioned above, also a probabilistic sensitivity analysis (PSA) was applied to evaluate the impact of simultaneous variation in model parameters on the model results. The results of the PSA were presented using a cost-effectiveness acceptability curve (CEAC) calculated from the net monetary benefit statistic across a range of willingness-to-pay (WTP) thresholds (Fenwick et al., 2006). The CEAC describes the probability that the “true” ICER estimate will be below the selected WTP. Furthermore, to study the total amount that the decision-makers should be willing to pay to eliminate all uncertainty in the decision, the expected value of perfect information (EVPI) was estimated as a function of WTP values (Wilson, 2015).

The expected costs and QALYs of all interventions are presented in Table 4. In the base-case, the use of the targeted PRS strategy improved the reclassification of individuals to the very-high risk category (i.e., a risk category receiving a lifestyle intervention and an additional medical therapy in the present study), on average, by 12.4 percentage points among individuals who would otherwise not been included in these activities using the usual practice. This, in turn, increased the number of years without T2D and T2D-related complications, on average, by 0.72% and 2.7%, respectively. In addition, in the long-term period, the use of the targeted PRS strategy decreased the risk of overall death by 2.0%, on average, as compared to the usual practice. These above changes resulted in cost savings, on average, −253€ per person as compared to the usual practice. In addition, the targeted PRS strategy produced, on average, 0.022 additional QALYs per person as compared to the usual practice. Thus, in the base-case, the targeted PRS-based strategy was a dominant intervention option with lower expected costs and a higher number of expected QALYs.

The results of one-way sensitivity analyses showed that the largest effect on the cost-utility results was the discount rate that was used with 0% discount resulting in cost savings of 444€ and 0.038 QALYs, whereas the use of 5% discount rate resulted savings of 165€ and 0.016 QALYs. Based on the results of the base-case analysis, the cost neutral point (i.e., ICER-value 0 €/QALY) was reached with a PRS test price of 303€. When shorter time horizons (i.e., 10- and 20-year time horizons) were applied the corresponding points were 19€ and 273€. The results for sensitivity analysis are shown in Table 5.

The probability of the cost-effectiveness of the PRS strategy remained stable as the WTP threshold increased (Figure 2). The probability of cost-effectiveness at a WTP of 0€/QALY was 63.0% indicating the probability that the PRS strategy is a dominant option (i.e., less costly, and more effective) conditional on the parameter uncertainty of the applied model. The corresponding EVPI estimate at a WTP of 0€/QALY was 243€ indicating the expected cost of uncertainty and the expected opportunity loss that could be avoided with perfect information.