Health economic analyses

Introduction

A mathematical model was constructed to estimate the changes in key outcomes following increases in the duration between monitoring period, from a standard 3-month period, for patients on anti-inflammatory drugs. The model focused on six drugs: 5-ASAs, anti-TNF alphas, leflunomide, methotrexate, sulfasalazine and thiopurine.

The outcome measures were:

• number of monitoring appointments (which decrease with increased monitoring intervals)
• number of additional AEs compared with standard monitoring (which increase with increased monitoring intervals)
• costs of monitoring appointments (which decrease with increased monitoring intervals)
• costs of additional AEs compared with standard monitoring (which increase with increased monitoring intervals)
• Quality-adjusted life-years (QALYs) gained (which decrease with increased monitoring intervals).

The purpose of the model was to estimate whether the costs saved through reduced number of monitoring appointments would outweigh the potential loss in health due to additional AEs and the costs of treating these AEs. This was determined by comparing the cost per QALY lost generated by the model with a standard National Institute for Health and Care Excellence (NICE) threshold of £20,000.370

The model provided incremental cost-effectiveness ratios (ICERs) expressed in terms of costs per QALY lost and the incremental net monetary benefit (INMB) for each drug for each monitoring period. INMB is defined as the incremental QALYs gained multiplied by the assumed cost per QALY threshold (£20,000) minus the incremental costs. Working in INMB has the advantage that strategies can be quickly compared with the highest INMB value being the strategy that is anticipated to be the most cost-effective at a chosen cost per QALY threshold. Advantages of NMB allows a comparison of multiple strategies without having to provide a matrix of ICERs and avoiding the use of a term such as dominated when a strategy costs marginally more (e.g. 1 pence more) and generates slightly less health (e.g. 0.0001 QALYs less). Here the INMB analysis would indicate the strategies were largely similar.

Within each analysis, ICERs were provided for patients who were representative of deciles of risk of having an abnormal blood test to explore whether better monitoring durations changed as this risk increased. These values are contained in appendices due to the number of results. However, a summary of the cost savings and QALYs lost at the lowest and highest risks is provided in the text, to allow the reader to assess the magnitude of these values.

Model overview

A health economic analysis plan was developed. This is available from co-author Matthew D Stevenson. The model used the statistical models described in Results to estimate the probability in each of the 20 quarters between the start of the model and year 5 that a patient would have an abnormal blood test that required cessation of the anti-inflammatory drug. With standard 3-monthly monitoring, patient’s cessation of treatment would occur at the relevant quarterly appointment. However, for longer monitoring intervals, it was assumed that the patient may have to wait longer (until the next appointment) for the anti-inflammatory treatment to be stopped. During this period, it is possible that AEs could occur that could harm the patient and result in expenditure. The potential AEs and the risks of each with continued use of an anti-inflammatory drug were chosen and estimated by clinical experts; the costs and QALY losses associated with each AE were sourced from reviews of NICE technology appraisals.

The risk of AEs due to increased intervals between monitoring appointments was based on clinical expert opinion as there are no relevant published studies to our knowledge. The risk of AEs was provided by Guruprasad Aithal (Hepatology), Christopher Fox (Haematology), Maarten Taal (Renal), Abhishek Abhishek (Rheumatology), Hywel C Williams (Dermatology), Timothy Card (Gastroenterology) and Christian Mallen (Primary care). To elicit the risk of AEs, the experts were provided with the task and requested to think about these risks based on their clinical experience. The experts convened in two online meetings lasting for 1–2 hours each. At these meetings, the risk was proposed, discussed, clarified and agreed. The agreed risks were e-mailed to the experts, giving them an opportunity to change their mind.

Monitoring appointment intervals of 3 months, 6 months, 1 year and 2 years were modelled. A time horizon of 5 years was chosen for the period to model monitoring appointments, although the potential health impacts of AEs were modelled over a patient lifetime. Discounting of future costs and future benefits was undertaken assuming a rate of 3.5% for each, as recommended by NICE.370 Patients were monitored according to each strategy for a period of 5 years (4 years in the biennial strategy although the impact of missed abnormal blood tests spanned the 5-year period). An additional monitoring appointment after cessation of treatment due to an abnormal blood test was assumed.

For each strategy, the costs associated with both monitoring and illness were estimated, as were the loss in QALYs. All values were discounted at a value of 3.5% per annum as recommended by the NICE.370 Results were presented in terms of INMB, assuming a cost per QALY threshold of £20,000, compared with monitoring every 3 months. Both probabilistic and deterministic analyses were performed. The strategy with the greatest INMB value is estimated to be the most cost-effective treatment. Sensitivity analysis considered threefold higher risks of illnesses than were estimated by the clinicians and a greater health and cost impact of AKI, and greater costs associated with sepsis. The model output was the bullet points reported in the Introduction. The model is not available publicly and is available from MDS.

Model parameterisation

Patient population

For methotrexate, thiopurine, sulfasalazine, 5-ASA and leflunomide, the generated risk prediction model was used to estimate the probability of outcome over a 5-year period in patient representative of each risk decile. These patients were selected by ranking the patients in terms of risk and selecting a random patient at the 5th percentile, 15th percentile, and so on until the 95th percentile (see Appendix 4, Tables 115–119). For anti-TNF alpha and mycophenolate, a prognostic model could not be constructed. Consequently, to account for uncertainty in outcome incidence, health economic analyses considered the midpoint and upper and lower 95% CI of the incidence rate from start of treatment, from 1 year after start of treatment, and within the first year of treatment. It was assumed that these risks were distributed exponentially. Probabilistic analyses were undertaken focusing on the data produced by the predictive model and assuming that other variables were fixed.

For each model, a new data set was produced using bootstrapping with a new model fitted to these data, providing different estimates of the time to an abnormal blood test and the coefficients for all relevant parameters. This was undertaken 1000 times for each of methotrexate, thiopurine, sulfasalazine, 5-ASA and leflunomide. These values could not be succinctly tabulated, and the data are not presented here.

Scenario analyses were undertaken to check the robustness of the results by combining the following: tripling the risks of each AE, increasing the costs associated with sepsis, and increasing the costs and QALY losses associated with AKI.

Risks of abnormal blood tests that lead to treatment discontinuation

The rates of discontinuation of anti-inflammatory treatment following an abnormal blood test were taken from the statistical models and used, for most drugs, for each of the 10 selected patients that represented deciles of risk. For anti-TNF alphas, a different approach was taken where nine risks of discontinuations due to abnormal blood tests were modelled based on the median and 95% CIs for the rate of discontinuation due to abnormal blood tests in the first 6 months, first year, and after the first year of prescription, which were 9.97 (7.34 to 13.54), 8.62 (6.74 to 11.01) and 3.44 (2.67 to 4.42) per 1000 person-years, respectively, estimated as described in Results.

The risk of adverse events if cessation of treatment is delayed due to longer intervals between monitoring appointments

The probability that an abnormal blood test would result in an illness because of the extended monitoring period was estimated by the clinical team members based on their experience. This involved a remote videoconference followed by asynchronous e-mail discussion.

The clinicians selected the potential AEs that could arise because the inflammatory treatment was not stopped at quarterly monitoring appointments but were stopped at the next appointment, which in the case of 2-yearly monitoring could be 1 year and 9 months longer than for quarterly monitoring. The AEs that the clinicians thought could occur were AKI; acute liver failure; anaemia; CKD; cirrhosis; DILI; early fibrosis; low neutrophil count plus sepsis; neutropenic sepsis; thrombocytopenia requiring hospitalisation; and thrombocytopenia with superficial bleeding. The clinicians estimated and erred towards overestimating the risks rather than underestimating them. increasing the risks. The risks associated with 6-monthly monitoring; yearly monitoring and 2-yearly monitoring are shown in Table 35.

For annual monitoring, the hepatologist and nephrologist in the team did not think the rates of clinically manifest hepatotoxicity and nephrotoxicity respectively would increase with annual monitoring, while the haematologist in the team thought the rates of clinically manifest haematological illnesses would double. The incidence of haematological side effects used in health economic analyses was doubled consequently. For biennial monitoring, the hepatologist in the team did not think the rates of clinically manifest hepatotoxicity would increase. The haematologist in the team thought the rates of clinically manifest haematological illnesses would quadruple. The incidence of haematological side effects was quadrupled consequently. The nephrologist in the team thought the rates of CKD would increase fourfold for other drugs and 8–10-fold for 5-ASA and sulfasalazine with biennial monitoring. The incidence of renal side effects used in health economic analyses was increased.

The assumed costs of monitoring blood tests

A monitoring appointment and blood tests were estimated to cost £24.09. The cost estimated for a monitoring appointment was based on 10 minutes of phlebotomist time (costed as band 4) at £5.83;371 5 minutes of GP time at £13.17 including qualification costs;371 £2.61 for a FBC, £1.24 for a LFT372 and £1.24 for a creatinine test (which was assumed to have the same price as a LFT, a renal function test and a calcium phosphate test).372 No disutility was assumed for attending a monitoring appointment.

The costs and quality-adjusted life-year losses associated with each adverse event

The costs and QALYs losses associated with each condition were estimated following targeted literature reviews (Table 36). All costs have been inflated to 2020–1 prices where required using inflation indices in Jones and Burns.371 The sources for these data have been selected from recent NICE appraisals where available assuming that NICE would have assessed these values, and from rapid searches where not. Where there was belief that the costs and QALY impacts could be underestimated, sensitivity analyses were undertaken increasing these values.

Acute kidney injury

The estimated cost for AKI used a weighted cost for the following NHS reference currency codes: LA07H, LA07J, LA07K, LA07L, LA07M, LA07N and LA07P. This provided a value of £1961, which was inflated to a value of £2022.373 Data on the utility or QALY loss associated with AKI were redacted in NICE TA755374 and so the QALY loss was estimated in an alternative manner. The number of days in hospital due to AKI was estimated by dividing £1961 by £378 (the average cost per regular day or night admission) which equalled 5.19. The QALY loss was assumed to be equivalent of 5 days with a loss of utility of 1, assuming that this captures QALY loss prior to hospitalisation and after discharge. This results in an estimated QALY loss of 0.01.375,376 A sensitivity analysis has been conducted assuming that the impacts from AKI are the same as for the pessimistic scenario for neutropenic sepsis as costs could extend beyond the hospitalisation period as could health burden on the patient.

Acute liver failure

The cost for acute liver failure in patients who were hospitalised was estimated using the following NHS reference currency codes: GC01C, GC01D, GC01E and GC01F, which resulted in a cost of £2461. The cost of liver transplantation was taken from GA15A assuming all patients were aged 18 years or over (£21,546), with ongoing annual maintenance costs after a liver transplant, assumed to last for a maximum of 20 years, taken from NICE TA755 (£5949).374 The proportion of patients who were hospitalised following ALF (53%), the proportion that require a liver transplant (8%) and those that died (5%) were taken from Andrade et al.377 Those who died or had a transplant were also assumed to incur the costs of an ALF hospitalisation stay. Taking the outcomes into account and inflating to 2020–1 prices, the weighted cost for patients having acute liver failure was estimated to be £3352. The estimated QALY loss associated with acute liver failure was calculated as the addition of the estimated losses associated with hospitalisation, transplant and death. For hospitalisation, it was assumed that the utility of a patient was reduced to zero for a period of a week. For transplant, it was assumed that the utility in the year of a transplant was 0.45 and then 0.67 in subsequent years as reported in NICE TA507,378 estimating a utility decrement compared with the average utility of a 60-year-old (0.816) of 0.366 in year 1, and 0.146 in subsequent years. It was assumed that this decrement would last for a period of 19 years.

For patients who die, the QALY loss associated with a death from sepsis reported in Stevenson et al.379 of 11.32 was used. Combining these components and considering the proportion of patients experiencing each event resulted in an estimated QALY loss of 0.651.

Anaemia

The costs for anaemia were taken from NICE TA507378 which provided a management cost of £369.57 and 4 weeks of erythropoietin treatment. When inflated to 2020–1 prices, the estimated cost is £465. For QALY losses, it was assumed that there would be the equivalent of half a day with a utility loss of 1.

Chronic kidney disease

The costs for a change in CKD stage were taken from NICE TA755374 assuming a change from CKD stage 3 to CKD stage 4, which reported an increase cost per year of £3030. It was assumed that the relative increase in stage would last for a period of 10 years, resulting in a discounted cost increase of £26,083. The utility loss per CKD stage was redacted in NICE TA755, but it was reported that these were similar to utilities reported in NICE TA599,380 which indicated that there was a 0.152 utility loss between CKD stage 4 and CKD stage 3. Assuming that the increase in stage would last for a period of 10 years results in a discounted QALY loss of 1.31.

Cirrhosis

NICE TA507378 reports that the combined initial evaluation costs and monitoring costs over the first 24 weeks were £2176 with annual costs of patients with compensated cirrhosis being £1138. Assuming that these increased costs were relevant for a 20-year period and inflating to a price year of 2020–1 results in increased costs of £21,700. The utility loss associated with cirrhosis was assumed to be 0.2 based on values used in NICE TA507378 which were 0.79 for non-cirrhotic patients with sustained virological response and 0.59 for patients with compensated cirrhosis with sustained virological response. Assuming a 0.2 utility decrement for a period of 20 years results in an estimated discounted QALY loss of 3.04.

In the absence of better evidence, it was assumed that the costs and QALY losses associated with DILIs were equal to those associated with ALF.

Early fibrosis

The utility loss associated with early fibrosis was taken from NICE Clinical Guideline 49,381 which suggested that a utility loss of 0.03 was associated with non-treated non-alcoholic steatohepatitis. This value was believed likely to be an overestimate in the public discussion of MRI-based technologies for the assessment of people with NAFLD.382 No costs were assumed to be associated with early fibrosis.

Low neutrophil count plus sepsis

The estimated cost for low neutrophil count plus sepsis was the weighted average of NHS reference costs associated with sepsis with no, or single, interventions (currency codes WJ06D, WJ06E, WJ06F, WJ06G, WJ06H and WJ06J). This was a value of £2313 when inflated to 2020–1 prices. The QALY loss associated with sepsis was comprised of a death component and a chronic disutility component. For death it was assumed, from Stevenson et al.379 that 13% of patients with sepsis would die, each losing 11.32 QALYs. For patients who did not die it was assumed that there would be a utility loss of 0.136, estimated by subtracting 0.68 (the utility post sepsis from Stevenson et al.) from the average utility of a 60-year-old (0.816). Assuming this disutility lasted for 10 years, there would be a discounted QALY loss of 1.02. Adding the utility losses, weighted by the probability of dying, resulted in an average QALY loss of 2.49 per person. A sensitivity analysis was conducted doubling the costs associated with neutropenic sepsis in case there were additional long-term costs post hospitalisation.

Neutropenic sepsis

The cost for neutropenic sepsis was the weighted average of NHS reference costs associated with sepsis with multiple interventions (currency codes WJ06A, WJ06B and WJ06C). This was a value of £8558 when inflated to 2020–1 prices. In addition, a cost of £898 was incurred for filgrastim use, resulting in a total cost of £9456. It was assumed that the QALY losses for neutropenic sepsis was the same as for low neutrophil count plus sepsis. A sensitivity analysis was conducted doubling the costs associated with neutropenic sepsis in case there were additional long-term costs post hospitalisation.

Thrombocytopenia requiring hospitalisation

The cost for thrombocytopenia was the weighted average of NHS reference costs associated with thrombocytopenia treatment (currency codes SA12G, SA12H, SA12J and SA12K) which was £771. Inflating this cost and adding £482 for a platelet infusion taken from Item Code BC045 from NHS Blood and Transplant prices assuming that two adult therapeutic doses were required resulted in an overall cost of £1276 [Ref price_list_bc_nhs_cost-per-item_2022-23_nocic.pdf (windows.net) accessed December 2022]. The QALY loss was assumed to be equivalent of 6 days with a loss of utility of 1, based on 6 days in hospitalisation using immune thrombocytopenic purpura as a proxy for thrombocytopenia requiring hospitalisation383 and assuming that this captures QALY losses prior to hospitalisation and after discharge. This results in an estimated QALY loss of 0.016.

Thrombocytopenia with superficial bleeding

This condition was assumed to be associated with an outpatient appointment cost, which was taken from NHS Reference costs and inflated to 2020–1 prices, resulting in a cost of £137.

Decision-analytic modelling results

Methotrexate

All extended monitoring periods were more cost-effective than 3-monthly monitoring and the INMB for each strategy is shown in Figure 10.

FIGURE 10 The INMB associated with extended monitoring periods compared to 3-monthly monitoring across deciles of predicted risk for methotrexate. X-axis shows increasing deciles of risk. (Figure not reproduced here; see the original article.)

For all deciles, monitoring every 2 years was estimated to be most cost-effective, but at higher risks the difference between annual and biennial monitoring was small (estimated to be an INMB of ˂ £25 per patient in the highest-risk decile). Probabilistic analyses provided results similar to the deterministic analyses and results were robust to changes in key parameters. A 2-yearly screening interval was estimated as being the most cost-effective strategy in over 100% of all simulation runs up to decile 9, assuming a cost per QALY threshold of £20,000. For decile 10, the probability of 2-year monitoring being most cost-effective was 99%, and the probability of annual monitoring being the most cost-effective was 1%. Appendix 5, Figure 60 shows the results when the risks of illnesses were assumed to be triple that estimated by the clinicians and the costs of sepsis were doubled and the costs and QALYs losses associated with AKI were set to that of neutropenic sepsis. In this sensitivity analysis, biennial monitoring was estimated to be most cost-effective up to decile 7, with annual monitoring most cost-effective for higher-risk deciles; all extended monitoring periods remained more cost-effective than 3-monthly monitoring. Disaggregated results for the base case are shown in Appendix 4, Table 120, with combined results in Appendix 4, Table 121. For an indication of the magnitude of the results, for patients in the first decile, moving to an annual monitoring period, rather than 3-monthly, resulted in a cost saving of £324 per person and a QALY loss of 0.0003; for those in the 10th decile, these values were £291 and 0.0011.

Thiopurine

All extended monitoring periods were more cost-effective than 3-monthly monitoring and the INMB for each strategy is shown in Figure 11.

FIGURE 11 The INMB associated with extended monitoring periods compared to 3-monthly monitoring across deciles of predicted risk for thiopurines. X-axis shows increasing deciles of risk. Reproduced with permission from Nakafero et al.1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

For all deciles, monitoring every 2 years was estimated to be most cost-effective, but at higher risks the difference between annual and biennial monitoring was small (estimated to be an INMB of ˂ £60 per patient in the highest-risk decile). Probabilistic analyses provided results similar to the deterministic analyses and results were robust to changes in key parameters. A 2-yearly screening interval was estimated as being the most cost-effective strategy in over 99% of all simulation runs up to decile 9, assuming a cost per QALY threshold of £20,000. However, the decision was more uncertain in decile 10 when the probability of 2-year monitoring being most cost-effective was 82%, and the probability of annual monitoring being the most cost-effective was 18%. Appendix 5, Figure 61 shows the results when the risks of illnesses were assumed to be triple that estimated by the clinicians and the costs of sepsis were doubled and the costs and QALYs losses associated with AKI was set to that of neutropenic sepsis. In this sensitivity analysis, biennial monitoring was estimated to be most cost-effective up to decile 7, with annual monitoring most cost-effective for higher-risk deciles; all extended monitoring periods remained more cost-effective than 3-monthly monitoring. Disaggregated results for the base case are shown in Appendix 4, Table 122, with combined results in Appendix 4, Table 123. For an indication of the magnitude of the results, for patients in the first decile, moving to an annual monitoring period rather than 3-monthly resulted in a cost saving of £325 per person and a QALY loss of 0.0003; for those in the 10th decile, these values were £295 and 0.0010.

Leflunomide

All extended monitoring periods were more cost-effective than 3-monthly monitoring and the INMB for each strategy is shown in Figure 12.

FIGURE 12 The INMB associated with extended monitoring periods compared to 3-monthly monitoring across deciles of predicted risk for leflunomide. X-axis shows increasing deciles of risk. (Figure not reproduced here; see the original article.)

Up to decile 7 (see Appendix 4, Table 124), monitoring every 2 years was estimated to be most cost-effective, but at any risk the difference between annual and biennial monitoring was small (estimated to be an INMB of ˂ £40 per patient in the lowest-risk decile). Probabilistic analyses provided results like the deterministic analyses and results were robust to changes in key parameters. A 2-yearly screening interval was estimated as being the most cost-effective strategy in over 97% of all simulation runs up to decile 5, assuming a cost per QALY threshold of £20,000. This proportion fell across deciles and was 93% for decile 6, 80% for decile 7, 52% for decile 8, 24% for decile 9 and 0% for decile 10, with annual monitoring being the compliment value. Appendix 5, Figure 62 shows the results when the risks of illnesses were assumed to be triple that estimated by the clinicians and the costs of sepsis were doubled and the costs and QALYs losses associated with AKI were set to that of neutropenic sepsis.

In this sensitivity analysis, biennial monitoring was not the most cost-effective strategy for any decile, with annual monitoring most cost-effective for all but the 10th decile where 6-monthly monitoring was marginally more cost-effective; 6- and 12-month extended monitoring periods remained more cost-effective than 3-monthly monitoring for all deciles, as was biennial monitoring for deciles 1–8. Disaggregated results for the base case are shown in Appendix 4, Table 124, with combined results in Appendix 4, Table 125. For an indication of the magnitude of the results, for patients in the first decile, moving to an annual monitoring period, rather than 3-monthly, resulted in a cost saving of £322 per person and a QALY loss of 0.0006; for those in the 10th decile, these values were £284 and 0.0020.

Sulfasalazine

Extended monitoring periods were more cost-effective than 3-monthly monitoring (Figure 13). For deciles one to nine, monitoring every 2 years was estimated to be most cost-effective, but at higher risks, the difference between annual and biennial monitoring was small (estimated to be an INMB of ˂ £20 per patient in decile nine). In the highest risk decile, annual monitoring had an INMB gain of ˂ £2 per person compared with biennial monitoring.

FIGURE 13 The INMB associated with extended monitoring periods compared to 3-monthly monitoring across deciles of predicted risk for sulfasalazine. X-axis shows increasing deciles of risk. (Figure not reproduced here; see the original article.)

Probabilistic analyses provided results like deterministic analyses and results were robust to changes in key parameters. A 2-yearly screening interval was estimated as being the most cost-effective strategy in over 98% of all simulation runs up to decile 8, assuming a cost per QALY threshold of £20,000. This proportion fell across deciles and was 80% for decile 9 and 31% for decile 10, with annual monitoring being the complement value. Appendix 5, Figure 63 shows the results when the risks of illnesses were assumed to be triple that estimated by the clinicians and the costs of sepsis were doubled and the costs and QALYs losses associated with AKI were set to that of neutropenic sepsis. In this sensitivity analysis, biennial monitoring was estimated to be most cost-effective up to decile 2, with annual monitoring most cost-effective for higher-risk deciles. Disaggregated results for the base case are shown in Appendix 4, Table 126, with combined results in Appendix 4, Table 127. For an indication of the magnitude of the results, for patients in the first decile, moving to an annual monitoring period, rather than 3-monthly, resulted in a cost saving of £329 per person and a QALY loss of 0.0001; for those in the 10th decile, these values were £302 and 0.0004.

5-aminosalicylate

All extended monitoring periods were more cost-effective than 3-monthly monitoring and the INMB for each strategy is shown in Figure 14.

FIGURE 14 The INMB associated with extended monitoring periods compared to 3-monthly monitoring across deciles of predicted risk for 5-ASA. X-axis shows increasing deciles of risk. (Figure not reproduced here; see the original article.)

For all deciles, monitoring every 2 years was estimated to be most cost-effective, but at higher risks the difference between annual and biennial monitoring was small (estimated to be an INMB of ˂ £25 per patient in the highest-risk decile). Probabilistic analyses provided results similar to deterministic analyses and results were robust to changes in key parameters. A 2-yearly screening interval was estimated as being the most cost-effective strategy in 100% of all simulation runs up to decile 9, assuming a cost per QALY threshold of £20,000. However, the decision was more uncertain in decile 10 when the probability of 2-year monitoring being most cost-effective was 92%, and the probability of annual monitoring being the most cost-effective was 8%.

Appendix 5, Figure 64 shows the results when the risks of illnesses were assumed to be triple that estimated by the clinicians and the costs of sepsis were doubled and the costs and QALYs losses associated with AKI were set to that of neutropenic sepsis. In this sensitivity analysis, biennial monitoring was estimated to be most cost-effective up to decile 7, with annual monitoring most cost-effective for higher-risk deciles, although the INMB difference between annual monitoring and biennial monitoring was ˂ £20 per person in deciles 1–9; all extended monitoring periods remained more cost-effective than 3-monthly monitoring. Disaggregated results for the base case are shown in Appendix 4, Table 128, with combined results in Appendix 4, Table 129. For an indication of the magnitude of the results, for patients in the first decile, moving to an annual monitoring period, rather than 3 monthly, resulted in a cost saving of £278 per person and a QALY loss of 0.0001; for those in the 10th decile, these values were £268 and 0.0002.

Anti-tumour necrosis factor

All extended monitoring periods were more cost-effective than 3-monthly monitoring and the INMB for each strategy is shown in Figure 15.

FIGURE 15 The INMB associated with extended monitoring periods compared to 3-monthly monitoring across deciles of predicted risk for anti-TNF-alpha. Reproduced with permission from Abhishek et al.4 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Not monitoring was most cost-effective. Among the monitoring strategies, monitoring every 2 years was estimated to be most cost-effective. Appendix 5, Figure 65 shows the results when the risks of illnesses were assumed to be triple that estimated by the clinicians and the costs of sepsis were doubled and the costs and QALYs losses associated with AKI were set to that of neutropenic sepsis. In this sensitivity analysis, not monitoring was estimated to be most cost-effective, with biennial and annual monitoring more cost-effective than 3-monthly monitoring or 6-monthly monitoring. Disaggregated results for the base case are shown in Appendix 4, Table 130, with combined results in Appendix 4, Table 131. For an indication of the magnitude of the results, for patients with the lowest risk, moving to an annual monitoring period rather than 3-monthly, resulted in a cost saving of £335 per person and a QALY loss of 0.0000; for those in the highest risk, these values were £329 and 0.0001.

Discussion

Our health economic analyses showed that undertaking monitoring blood tests once every 6 months, annually and biennially were all more cost-effective in most deciles of risk than undertaking these tests at 3-monthly intervals and even when worst-case scenarios were assumed. Lengthening the time between monitoring blood tests would save patient and clinician time, minimise discomfort from unnecessary venepunctures and conserve healthcare resources.

There are several strengths of this study. First, the health economic analyses used real patient scenarios across the entire risk profile of potentially eligible patients in the UK, thus increasing generalisability of the findings. We used a large, real-world and nationally representative data set for selecting scenarios. Second, the study population included patients with a range of IMIDs, and the results have broad generalisability. Third, the monitoring strategies that were most cost-effective were robust to changes in assumptions, for example, related to threefold higher probability of illnesses following delayed abnormal blood test. Limitations included lack of published data on the relationship between abnormal blood-test results and clinically significant illnesses, and the rates of the latter were estimated by clinicians. While the experts presented their interpretation of the risk of AEs, this was based on their clinical opinion and could be subjected to personal biases. The health economic modelling considered tripling of risk to account for clinician underestimation of risk.

In conclusion, we undertook health economic analysis that showed cost-effectiveness of less frequent monitoring in patients treated with methotrexate. These findings need to be considered by national and international specialist societies’ guideline writing groups to decide upon frequency of monitoring blood tests during long-term treatment.