Potential cost-effectiveness of childhood screening for type 1 diabetes: Examining reducing Diabetic Ketoacidosis at diagnosis and long-term outcomes to assess the justifiable cost per child screened.

Patient and Public Involvement

A PPI group of parents (1 one whom did not have experience of T1D) was established prior to submitting the bid, and the group was involved throughout. The parent of a child diagnosed with T1D was a full member of the project management group and contributed to the design of the economic analyses. The results of the economic analysis were then presented to a group of parents with T1D.

Health economic analysis plan

A health economic analysis plan was not developed for this work.

Population

Our population was the general population of children aged 4 years because we believed that this would be the optimal age for a single screen test, and if implemented, this could be done at the same time as the pre-school vaccination in the UK. The general population comprised 50% males. The population diagnosed with T1D was assumed to be 52% male based on National Paediatric Diabetes Audit (NPDA)¹³.

Setting, perspective, discounting and time horizon

Our study used the English and Welsh NHS setting. Consequently, we followed the National Institute for Health and Care Excellence (NICE) guidelines and took an NHS and personal social services perspective, discounted future costs and QALYs at 3.5% per annum, and took a lifetime horizon¹⁴.

Intervention

The intervention considered was a screening test to detect T1D in every child in England and Wales, aged 4. To detect T1D in the general population, we assumed that the screening program would perform identically to the screening program presented by Ziegler et al. in Germany⁸.

Comparator

The comparator was no screening for T1D (current practice).

Selection, measurement and valuation of outcomes

QALYs were the key measures of benefits and harm. The key benefits of screening included prevention of DKA at diagnosis and potential long-term benefits from preventing DKA at diagnosis in scenario analyses.

For our base case, we applied a hazard ratio to reduce the probability of DKA presentation compared with the control arm.

Utilities for health states in the model were sourced from the literature with values based on the EQ-5D self-reporting by populations of interest, valued using the England and Wales EQ5D3L value set, which were preferred in line with NICE guidance^15–17.

Currency, price data and conversion

All costs in the economic model are in pounds sterling and are 2020/21 prices. The costs from previous price years were inflated to 2020/21 prices using the NHSCII inflation index¹⁸.

Rationale and description of model logic

Figure 1 shows the decision-tree structure for a single strategy (screening or no screening). The population enters the model and is assigned as either truly having T1D or not having T1D. For the population subgroup without T1D, their quality-adjusted life expectancy was estimated using Office for National Statistics (ONS) lifetables to estimate their life expectancy and age-gender norms for utility^19,20. For the population with T1D, we assumed for simplicity that they developed T1D immediately. The population subgroup with T1D was split into two further subgroups, one who developed DKA at the time of diagnosis and the other who did not. In the screening arm, a lower probability of having DKA is applied than in the no-screening arm; in the base case, this is the only benefit of screening. Then, the proportion of the population present in the DKA is assumed to be present immediately and can either survive or die from the DKA. Therefore, fewer children with T1D will die if they undergo screening, as fewer of these children have DKA. As they develop T1D immediately, this means that our estimates will be slightly favorable to screening, as the costs and utility decrements due to DKA will not be discounted because they occur in the future. The probability of death is taken from the NPDA in the base case. For the proportion of the population that has a DKA at diagnosis, every child was assumed to have an admission for their DKA, and an associated cost was added to their overall costs of treatment¹⁹. For the proportion of the population that has T1D and survives (either with or without an initial DKA at diagnosis), their estimated lifetime costs and QALYs are estimated using a modified version of our individual-level simulation model designed to estimate outcomes for adults with T1D²¹. In the base case, these projections of lifetime costs and QALYs are the same in both the screening and no screening arms. Scenarios of long-term benefits of HbA1c, severe hypoglycemia, and future DKA (which will change the projected lifetime costs and QALYs for children with T1D) were explored in scenario analyses.

Figure 1. The model diagram for a single strategy (either screening or no screening).

In a decision tree, the expected probability that each outcome occurs is multiplied by the expected costs and QALYs of that outcome to generate the expected outcome. All expected outcomes are added together to provide the overall expected costs and QALYs.

Analytics, assumptions and study parameters

Clinical probabilities

A summary of the clinical probabilities used in the economic model is presented in Table 1. Detailed descriptions of the evidence base are provided in the subsequent sections.

Probability of having T1D

We used the study by Zeigler et al., which was a prospective cohort study of screening in the general population of children aged 1.75–6 years in Bavaria, Germany⁸. They found that 0.31% of their population developed T1D. We assumed that this percentage would be relevant for screening individuals aged 4 years in England and Wales. This is based on age 4 being approximately in the middle of this age range, clinical advice, and the fact that the study was conducted in Germany, which we would expect to be similar to an English and Welsh setting⁸.

Probability of having a DKA at diagnosis, if the child has T1D

Based on the most recent NPDA data for 2019/20, we assumed that in the absence of screening, 38.5% of children with T1D would have a DKA event at their diagnosis²². For children who were screened, we assumed that there would be an 80% reduction in their risk of having a DKA at diagnosis (this is implemented as a Hazard Ratio of 0.2, as we expected this evidence to be generated in a time-to-event analysis). This was sourced from our early work that performed power calculations for subsequent randomized controlled trials of screening, with DKA prevention as the primary endpoint. This was based on the observed effectiveness. This gave the probability that a child who was screened and developed T1D would have DKA at a diagnosis of 9.3%.

Probability of dying from DKA at diagnosis

From the most recent NPDA report on childhood outcomes, there were four deaths related to DKA (but not specifically DKA at diagnosis) in children aged 0–14 between 2015–20²³. In the NPDA, 14,363 DKA events were detected over this period²². This gives a 0.03% conditional probability that a child, given that they have a DKA at diagnosis, could die from this event.

Decision tree outcomes

People without T1D

For people without T1D, we estimated their life expectancy from the Office for National Statistics (ONS) life table mortality data¹⁹. To get their utilities, we used values reported by the NICE decision support unit (DSU) by age and gender for the English and Welsh population²⁰. As the NICE DSU data is only valid for adults, we assumed that all children would have the same utility as a 16 year old from this analysis. These data were combined to estimate the expected QALYs of this population. This gave a predicted life expectancy of 77.30 years, when applying utilities were applied, a total quality-adjusted life expectancy of 67.02 QALYs, when discounted, was applied at 3.5% per annum, which gave a discounted quality-adjusted life expectancy of 24.61 QALYs.

People with T1D, who die from a DKA at diagnosis

We assumed that the death caused by DKA occurred on the day of diagnosis. Therefore, we included people who died from DKA at a diagnosis of 0 QALYs.

This incurs a treatment cost for the treatment of DKA. This was £3,278, which was obtained from the National Schedule of Reference Costs²⁴, Currency Code PK67A (Paediatric Diabetes Mellitus, with Ketoacidosis or Coma, with CC Score 1+) and PK67B (Paediatric Diabetes Mellitus, with Ketoacidosis or Coma, with CC Score 0). This was estimated as a weighted average from all the times these currency codes were used in 2020–21.

People with T1D who do not die from DKA at diagnosis

To reflect the expected costs and QALY outcomes of our population we conducted a run of our Sheffield Type 1 Diabetes Policy model²¹. This model was run using 2015/16 prices and then the total discounted costs over an average child with T1D’s life expectancy was inflated to 2020/21 prices.

In this analysis, we used our baseline characteristics from the NICE guidelines for modelling CGM in adults and did not allow any risk factors to change over time. We set the age to 4 and the proportion of males to 52% based on NPDA data:

The probability of having subsequent DKAs and severe hypoglycemia was taken from NPDA data on the probability of having a DKA or severe hypoglycemic event by diabetes duration²². We assumed that one event occurred per person per year, due to the absence of information on the number of events that occurred per year. We assumed that the probability of having DKAs in the group with the highest diabetes duration in the dataset was maintained for the rest of the patient’s lifetime.

We adapted our model so that simulated individuals could not develop diabetes-related complications (cardiovascular disease, retinopathy, macular edema, neuropathy, or nephropathy) until they were 18. The model estimated lifetime costs and QALYs for this population. We took annual costs for managing T1D from evidence submitted to NICE as part of their appraisal of hybrid closed-loop technology (TA943, £5,387per year on insulin and devices and £575 per year for contacts with health care professionals)²⁵.

In addition, the cost of presenting with DKA at diagnosis and a utility decrement associated with DKA (for children who survived their DKA at diagnosis) were also included, which was done in our decision tree and not in the long-term model being used to generate quality-adjusted life expectancy. A summary of the expected lifetime discounted costs and QALYs for these children with T1D are given in Table 2 and a summary of the baseline characteristics used in the Sheffield T1D model are given in Table 3.

People with T1D

The assumptions on costs and QALYs incurred by this population group were the same as those of the population with T1D who did not die from DKA at diagnosis; however, they did not incur the cost of treating their DKA at diagnosis or the associated utility decrement.

Economic analysis

We ran our analysis without any assumed costs for screening, as it is too early at this stage to estimate the cost of a general population screening program for T1D. What we did Instead, we calculated the cost of screening to provide an Incremental Cost-Effectiveness Ratio (ICER) of £20,000 per QALY gained. The ICER is calculated as:

Where the cost of screening itself is not included in the screening costs

We rearranged this to give:

We used an ICER of £20,000 per QALY gained, as this is usually the lower end of the range of ICERs that the NICE considers acceptable. This provides an upper limit on the cost of screening, given the expected costs associated with T1D and the associated QALYs of people being screened.

Characterizing heterogeneity

No subgroup analyses were explored

Characterising distributional effects

No adjustment was made to reflect the priority populations, as this study concerns the feasibility of general population screening.

Sensitivity of results to effect size on primary outcome

As we were developing an early model to identify key drivers for an economic screening case, we did not include any probabilistic analyses that would be expected in any economic evaluation of an actual screening program. The primary reason for this is that the most important uncertainty relates to the effectiveness of the screening program in reducing the number of people whose first knowledge of T1D is being diagnosed as an emergency in a hospital with a DKA. We do not yet have data on the effectiveness of an English-and-Welsh screening program and the statistical uncertainty in this effectiveness. We conducted a series of scenario analyses to explore the uncertainty in the true values of the model inputs.

Sensitivity of results to other parameters

We conducted 15 further exploratory analyses to explore the scale of the impact of other potential effects of a screening program beyond the DKA at diagnosis. The exploratory analyses were as follows:

A summary of the discounted quality-adjusted life expectancy and costs for children with T1D across our conservative and exploratory analyses is presented in Table 4.

Ethics

Ethics approval was not required for this research, as all data were obtained from the published literature.

Summary of the main results

Our results, using only the clinical data that would be generated from an RCT of screening for T1D designed to detect a difference in the incidence of DKA at diagnosis, are given in Table 5. This shows that in this analysis, we would need the cost of screening to be under £3.17 per child screened (i.e., including the costs for children who do not develop T1D) for a screening program to have an ICER of £20,000 per QALY gained. This is an important threshold for an ICER, as this is the lower end of what NICE usually uses to determine the cost-effectiveness of health technologies and strategies in the English and Welsh NHS, and is also used by the National Screening Committee.

Effect of uncertainty

Our scenario analyses, which explore whether other benefits of screening for children with T1D can be proven, are given in Table 6. Table 6 shows that if we can find other substantial benefits (other than reducing the risk of DKA at diagnosis) from screening for children with T1D, the maximum acceptable cost of screening and follow-up could increase to approximately £50 per child screened. Other scenarios with large impacts on cost-effectiveness include setting the population to all have T1D, increasing the probability of death from DKA, and assuming that all deaths in patients under 14 years of age due to endocrine disorders were due to DKA at diagnosis.

Strengths and limitations

The key strength of this study is that it addresses the screening costs in England and Wales in various scenarios. This gives us a good idea of what screening could cost in the UK, and what would be the key clinical factors if demonstrated to be the most likely scenario when screening.

Through the scenarios on screening affecting long-term biomedical outcomes, this not only covers scenarios where metabolic control is improved, regardless of the cause. This could be related to preventing DKA at diagnosis or by preventing the psychological trauma of an acute and life-threatening DKA in a previously healthy child. By calculating the mean cost, we could not specify any particular part of the screening design. Therefore, the work is translatable to a variety of potential screening programs with different invitation strategies, tests, follow-up testing schedules, and support programs or interventions to reduce the incidence of DKA for parents and children who test positive for T1D (we do expect that there will be some false positives).

As this was an early analysis, there are many limitations that should be considered when interpreting the results. The key limitations of our study are as follows.

This study does not include the costs or effects of teplizumab, an agent for delaying the onset of stage 3 T1D which is currently undergoing assessment by NICE in a population aged over 8 with stage 2 T1D³⁷. We did not include this, firstly we do expect that many children that would be identified in a screening programme for 4 year olds would remain free of T1D until they were 8, therefore the effects of adding teplizumab are likely to be small.

We have assumed that a screening programme in the UK would achieve exactly the same outcomes as in Ziegler et al.⁸. A UK based programme may not have exactly the same outcomes as those in Germany. This means that our current model does not have a natural history component modelling of undiagnosed stage 1, stage 2, and stage 3 T1D in the general population. This implies that we cannot explore different tests for detecting T1D in our current modelling framework.

We have not conducted any primary research; as such, we do not include any utility decrements or increased rates of health care contacts related to psychological distress related to either a child presenting in DKA (to either the parent or the child), or psychological distress related to screening itself. These effects, if shown to exist, should be factored into any cost-effectiveness analysis of screening for type 1 diabetes.

There may be other economic benefits accrued from any health care program that are outside the NHS and personal social services perspective for economic evaluations set in England and Wales (e.g., preventing time off work for parents as their child is less likely to be present in DKA); however, what is not clear with these benefits is who should pay to gain these benefits (e.g., the parents themselves, their employers, or the government). There may be other economic benefits that are not included because of the absence of evidence. For example, we may be able to evaluate any changes in the psychological burden on the parent/caregiver and the patient.

Because we adopted a decision tree analysis, screening was performed in a snapshot. An actual screening program will have complexities (e.g., time from screening until diagnosis of T1D and testing schedule) that would require a type of model that explicitly includes time in the structure. This may cause changes in the cost and QALYs associated with screening. However, we expect these to be relatively small changes, and we would require extensive epidemiological studies on screening children for T1D in the UK, which currently does not exist.

Our estimates of the expected costs of treating children with T1D are likely to be overestimated. In the NICE appraisal of hybrid closed loops, evidence submitted to this appraisal was used as some of our key costing information for device and treatment costs for T1D, the committee’s preferred base case ICER was £104,003 per QALY gained and their acceptable ICER was £20,000 per QALY gained³⁸. As hybrid closed loop is an approved technology in the UK, it is likely that there are commercially confidential arrangements between the device manufacturers and the UK government that effectively lowers the price of HCL from the publicly available prices we used when estimating the discounted lifetime cost of having T1D.

Future research

The key area for future research is to establish whether preventing DKA at diagnosis leads to long-term benefits in children with T1D. This is because it is unlikely that we could design a screening program that costs only an average of £3 per child screened. If these benefits exist, we have a much better chance of designing a screening program that can cost more than £20 per child screened, and if these benefits can be reasonably assumed to be permanent, then the cost per child screened could rise to over £50 per child screened.

Our model was moderately sensitive to increasing the number of deaths due to DKA at diagnosis (the cost of screening increased from £3.17 to £7.28), so obtaining better information on this model parameter would be useful.

It would be useful to develop an economic model of T1D that appropriately includes the incidence of diabetes-related complications during childhood. Our approach in this project of not allowing diabetes-related complications was simple and may have been inaccurate. However, this gap will likely require the development of new epidemiological risk equations for diabetes-related complications in T1D children.

If teplizumab or other agents are approved to delay T1D in children, the available information should be included in this analysis. However, if these are completely new products to the UK market, it is likely that they will be priced to capture the benefits obtained by using the treatments. This impact is likely to be small in terms of the cost-effectiveness of screening T1D. However, they could be essential for designing a screening program for T1D, as they would give children who test positive for treatment. This would likely increase the acceptability of screening for T1D among parents.

Other literature

There have been two previous analyses of T1D screening in children. There was an analysis by Karl et al. that conducted an analysis of the costs of screening in the Fr1da study by statistically analysing cost data collected in FRIDA and from routine sources in Germany³⁹. This was a within study costing of the Fr1da data and they found that screening for T1D as implemented in Fr1da cost €28.17 per child screened. There will be some differences between the German and English/Welsh settings of the Fr1da study and our analysis; however, if we can establish that there are some long-term benefits of screening, then it may be cost-effective to screen young children (four years old) for T1D. There was an analysis by McQueen et al. In the US setting, McQueen et al. found that avoiding DKA events alone would not result in cost-effective in the US setting. The model they used was an initial state transition up to age 30, which only tracked if children developed T1D and if they died, and a lifetime simulation model after age 30 to track the incidence of complications associated with T1D over children’s lifetimes. This is similar to our result, albeit with the US having a higher maximum acceptable ICER generally than England and Wales, they found that screening could be cost-effective with only an HbA1c improvement of 0.1% over a lifetime⁴⁰. This again, is similar to our results. A recent review by Rewers, found similar results in that DKA prevention alone would not be sufficient to make screening for T1D worthwhile⁴¹. However, Rewers also made a point that societal costs, such as productivity and quality of life of the parents when their child is diagnosed would be additional economic benefits to those gained directly by the children with T1D with regards to their health⁴¹.