The secondary outcomes addressed in this chapter are:
- Secondary outcome 3:
- Average per-patient national healthcare cumulative costs incurred within 5 years of newly diagnosed CD.
- Secondary outcome 4:
- Patient, disease phenotype and imaging characteristics associated with higher health economic costs, within 5 years of diagnosis.
Methods
Overview of cost evaluation
All costs are calculated from an NHS perspective, specifically, including data on hospital resource use.
We estimated the healthcare costs incurred within 5 years of a new diagnosis of CD and investigated patient, imaging, treatment and other factors that drive these costs. We collected NHS hospital resource use data for all patients over the 5-year follow-up period and applied unit costs to these data to calculate average (mean and median) 5-year costs per patient. We used regression analysis to identify the patient, imaging and disease characteristics driving higher NHS costs. This provides valuable information regarding the costs of CD and the factors associated with these costs, which are data that are scarce currently for the NHS setting, particularly in an era where biologic therapy is administered earlier in the disease course. Mean costs per patient were also multiplied by the estimated number of UK patients, stratifying by presence or absence of disabling disease within 5 years, to estimate the UK cost of illness (COI) following diagnosis.
Cost analysis took a UK NHS and Personal Social Services perspective.68 Costs were calculated in 2020–1 Great British pounds (£) and the time horizon was 5 years. Discounting was not applied to costs or outcomes, as this is a retrospective costing analysis and not a decision analytic model. Hence, no consideration of preferences for present benefits over future benefits is required.
Baseline patient, imaging, treatment and other factors
The baseline factors considered in the regression analyses are those from the primary modelling analysis (see Chapter 2) and include the nine prespecified clinical factors at diagnosis (unless otherwise specified): age, smoking status, presence of stricturing, penetrating disease or perianal disease, initial need for steroid therapy, sex, weight loss of at least 5 kg prior to diagnosis, severe endoscopic disease and location of disease. Other factors include the MRE imaging scores (LI, MEGS and sMARIA, each standardised to a 0–100 scale) and the development or not of severe/disabling disease within 5 years, using the modified Beaugerie criteria.
Cost components
Resource use data were obtained on the following cost components for every patient in the study:
- Medications the patient received in secondary care for CD (name, dose, frequency, start date and either confirmation that the drug was ongoing at the follow-up date or the end date or the duration for which the drug was taken).
- Imaging and endoscopic assessments related to CD, to include numbers of tests and dates of examination at diagnosis and during the follow-up period (e.g. colonoscopy, flexible sigmoidoscopy, and so on.).
- Surgical treatments related to CD (procedure, whether day case or not, whether elective, date of procedure and date of discharge, and total length of stay).
- Total number of outpatient visits and non-surgical day case visits related to CD over the follow-up period. Some patients had longer follow-up periods than 5 years (up to 12 years), but since we did not capture dates for these outpatient and day case visits, we instead included a pro rata proportion of the total number of visits to cover the 5-year period using the equation: number of visits over 5 years = number of visits over X years * 5/X, assuming a uniform distribution of events in accordance with discussions with clinical colleagues.
- Inpatient stays related to CD, excluding those related to surgery recorded in item 3 above (reason for stay, specialty, dates admitted and discharged, and total length of stay).
Uniform distribution of outpatient and non-surgical day case visits
Regarding assumption 4, it was felt by the clinicians that some patients could have more appointments earlier on in their disease course and then fewer once their disease had stabilised. Conversely, other patients may have had stable CD to begin with but then subsequently have more complex disease requiring more healthcare resource input. Indeed, other patients could fall between these two extremes and have a fairly uniform distribution of appointments across their post-diagnosis period. After discussion with clinical members of the trial group, it was agreed that on average the distribution of Outpatient and day case events would be more or less uniform and there was no clear evidence towards these events being more frequent either early or late within the 5-year period to provide meaningful input parameters for sensitivity analysis. Any impact of modifying these relatively low-cost appointments would in any case have been minimal, as they accounted for only around 16% of the total cost.
Unit costs
The CRF requested information on specific scans and other assessments for imaging and endoscopy assessments. The NHS Reference Costs include costs for these assessments on various tabs (including Outpatient Procedures, Imaging, Nuclear Medicine), and unit costs were assigned on the basis of discussions with clinical colleagues according to what was the most relevant setting in this patient group in order to allow more accurate and appropriate costs.
Unit costs (Table 17) for imaging and endoscopic assessments were taken from the Outpatient Procedures tab (colonoscopy (FE32Z), endoscopy (FE50A), gastroscopy (FE22Z), sigmoidoscopy (FE35Z)), the Imaging tab (dexa scan (RD50Z), fluoroscopy (RD32Z), magnetic resonance cholangiopancreatography [(MRCP) RD01A], MRI (RD05Z), MRI enteroclysis (RD05Z + RD30Z), US (RD40Z), X-ray (PF), CT scans (RD24Z), barium follow-through (RD32Z) and contrast enema (RD30Z)), and the Nuclear Medicine tab [positron emission tomography (PET), RN07A] and PET-CT scans (RN01A)) of the 2020–1 NHS Reference Costs.69
| Imaging and outpatient procedures | Unit costs (£) |
|---|---|
| X-ray | 57.78 |
| US | 106.26 |
| DEXA scan | 131.18 |
| CT scan | 152.64 |
| Contrast enema | 187.52 |
| Endoscopy | 189.37 |
| MRCP | 211.24 |
| MRI | 245.37 |
| Fluoroscopy | 336.59 |
| Barium follow-through | 336.59 |
| Colonoscopy | 345.55 |
| MRI enteroclysis | 432.89 |
| Sigmoidoscopy | 499.32 |
| Gastroscopy | 615.22 |
| PET scan | 697.61 |
| PET-CT scan | 1127.81 |
Unit costs for outpatient visits (£168.02) used weighted means from General Surgery, General Medicine, Gastroenterology and Dietetics from the Total Outpatient Attendances tab. Unit costs for non-surgical day case visits (£632.98) used weighted means from the Day Case tab for Non-Malignant GI Tract Disorders (FD10D/F/G/H/J/K/L/M).
Unit costs for inpatient stays where accident and emergency (A&E) attendance was reported were costed as ‘A&E not admitted’ if length of stay was zero days, as ‘A&E admitted plus non-elective short stay’ if length of stay was more than zero days, and as ‘Emergency Medicine, Category 1 Investigation with Category 1–2 Treatment’ in either case. Unit costs for planned inpatient stays of shorter than 3 days were costed as the weighted mean of ‘Day Case, IBD without interventions’ if the specialty was indicated as General Medicine or Gastroenterology, and as the same but with a single intervention if the specialty was indicated as Surgery. Unit costs for planned inpatient stays of 3 days or longer were costed as the weighted mean of ‘Elective Inpatient, IBD without interventions’ if the specialty was indicated as General Medicine or Gastroenterology, and as the same but with a single intervention if the specialty was indicated as Surgery. Trim points and excess bed day costs from the 2023–4 NHS Reference Costs were included to allow additional cost for the longer stays (Table 18).
| NHS Reference Cost category and assumptions | Unit cost (£) | Specialty according to study data | Which page on CRF |
|---|---|---|---|
| AE not admitted | 124.00 | A&E | Page 8 |
| AE admitted + NES IBD weighted mean no interventions | 867.30 | A&E plus NES stay | Page 8 |
| DC IBD weighted mean no interventions | 388.94 | General medicine/gastroenterology | Page 8 |
| DC IBD weighted mean single intervention | 2493.69 | Surgery | Page 6 and 8 |
| EL IBD weighted mean no interventions (trim point 5 days, excess bed-day cost £298) | 1780.48 | General medicine/gastroenterology/haematology | Page 8 |
| EL IBD weighted mean single intervention (trim point 17 days, excess bed-day cost £298) | 6025.45 | Surgery | Page 6 and 8 |
| NES IBD weighted mean single intervention | 741.30 | Surgery | Page 6 |
| NEL IBD weighted mean single intervention (trim point 17 days, excess bed-day cost £298) | 4991.87 | Surgery | Page 6 |
Unit costs for surgeries were costed as ‘day case, elective’ or ‘Non-elective short’ (length of stay < 3 days) or ‘non-elective long stay’ (NEL) (more than 3 days), according to those variables (see Table 18) using similar methods as those described above for inpatient stays.
Unit costs were multiplied by resource use collected in the trial. For CD medication costs, we used the cheapest available Drug Tariff cost listed in the online British National Formulary unless a specific brand name was specified, in which case that brand was costed. If drug tariff prices were not available and a brand name was not specified, we used the lowest available NHS Indicative Price instead. Costs of whole packets were used. There were very little missing data, which were left as missing unless there was an obvious suggestion in another free text variable regarding reasonable or commonly indicated doses or routes of administration to be imputed.
Patient-level data capture and time horizon
No directly identifiable patient information was used for this analysis.
Data were included for costing from the date of diagnosis and for the following 5 years. Patients with shorter than 5 years of follow-up were omitted. Patients with longer than 5 years of follow-up had their resource use data truncated at the 5-year point. Information from before the diagnosis date was also removed.
Medications and other treatments, procedures and inpatient stays that began before the date of diagnosis were truncated at the date of diagnosis (i.e. exclude earlier period), and those that continued after the 5-year date were also truncated (i.e. exclude later period).
Regression analysis
Regression analysis methods were used to investigate patient, disease phenotype and imaging characteristics associated with higher health economic costs within 5 years of diagnosis. Unadjusted 5-year costs were calculated separately according to the presence versus absence of severe/disabling CD, treatments received and patient demographic characteristics.
Multivariable regression was used to identify factors (CD status, MRE parameters, treatments received, patient characteristics) associated with higher costs, in accordance with the choice of variables from the risk model.
To account for skewed cost data, a generalised linear model with gamma family and log-link was used, experimenting with other distributional assumptions (log-normal, Gaussian, inverse Gaussian and negative binomial distributions), and the best fit was selected as judged by residual plots and the Akaike information criterion (AIC). The BIC was also examined to assess agreement between AIC and BIC regarding the best-fit model and the most statistically important cost drivers, although the BIC can have a tendency to under-fit.
Calculating the cost of illness in recently diagnosed patients
Mean costs per patient were multiplied by the estimated number of patients in the UK with CD, stratifying by presence or absence of disabling disease according to the proportions of these two groups calculated during our analysis, to estimate the COI for the 5 years following diagnosis in the UK. COI studies can be based on prevalence or incidence, depending on the research question and data availability. Prevalence-based approaches allow calculation of costs over a fixed period of time, for example, 5 years, according to the number of people in the region or country known to currently be suffering from the disease. Incidence-based approaches instead aim to calculate lifetime costs for patients, using yearly incidence rates to provide numbers of new patients joining the population of interest each year.70 In the context of this analysis, we have calculated the 5-year prevalence-based COI, and focus on those patients who are within the first 5 years of their diagnosis.
Results
Resource use and costs covering 5 years from diagnosis
Table 19 shows unadjusted mean (standard deviation) and median (IQR) total per-patient costs of health care for the categories of costs as collected in the study CRF. Mean (SD) total costs per patient were £24,267 (£33,108) for the whole group. When split according to whether they had developed severe/disabling disease by 5 years, according to the modified Beaugerie criteria, costs were £29,763 (SD £38,278) for those with disabling disease and £20,327 (SD £28,368) for those without, over the 5-year period.
| n (%) | Mean | SD | Median | IQR – lower | IQR – upper | ||
|---|---|---|---|---|---|---|---|
| All patients at 5 years | |||||||
| Total unadjusted costs over 5 years from diagnosis, £ | |||||||
| Drug costs | 146 | (75) | 21,592 | 25,741 | 8988 | 913 | 37,844 |
| Drug costs – biologics | 87 | (45) | 34,051 | 25,568 | 32,961 | 13,271 | 41,553 |
| Drug costs – non biologics | 143 | (74) | 1328 | 1905 | 562 | 225 | 1794 |
| Inpatient stays | 62 | (32) | 9288 | 27,765 | 4500 | 2078 | 7481 |
| Surgeries | 41 | (21) | 8999 | 10,238 | 6025 | 4992 | 8227 |
| Non-surgical day cases | 30 | (15) | 1860 | 3487 | 643 | 482 | 1130 |
| Outpatient visits | 150 | (77) | 2124 | 941 | 1997 | 1516 | 2666 |
| Imaging | 151 | (78) | 1564 | 1063 | 1427 | 736 | 2116 |
| Total costs | 194 | (100) | 24,267 | 33,108 | 10,136 | 1856 | 41,092 |
| Patients with disabling disease at 5 years (modified Beaugerie) | |||||||
| Total unadjusted costs over 5 years from diagnosis, £ | |||||||
| Drug costs | 57 | (70) | 25,429 | 23,718 | 20,220 | 3323 | 38,007 |
| Drug costs – biologics | 43 | (53) | 31,566 | 22,657 | 32,093 | 12,677 | 40,842 |
| Drug costs – non biologics | 56 | (69) | 1645 | 2438 | 773 | 280 | 2028 |
| Inpatient stays | 43 | (53) | 10,994 | 33,138 | 4553 | 2078 | 7481 |
| Surgeries | 25 | (31) | 8896 | 9919 | 6025 | 6025 | 8227 |
| Non-surgical day cases | 9 | (11) | 950 | 695 | 581 | 547 | 1382 |
| Outpatient visits | 58 | (72) | 2490 | 918 | 2451 | 1886 | 2891 |
| Imaging | 59 | (73) | 1919 | 1112 | 1646 | 1095 | 2402 |
| Total costs | 81 | (100) | 29,763 | 38,278 | 17,352 | .0 | 45,969 |
| Patients without disabling disease at 5 years (modified Beaugerie) | |||||||
| Total unadjusted costs over 5 years from diagnosis, £ | |||||||
| Drug costs | 89 | (79) | 19,135 | 26,798 | 3620 | 555 | 33,307 |
| Drug costs – biologics | 44 | (39) | 36,480 | 28,175 | 33,640 | 19,015 | 42,728 |
| Drug costs – not biologics | 87 | (77) | 1125 | 1444 | 453 | 176 | 1292 |
| Inpatient stays | 19 | (17) | 5427 | 5236 | 2922 | 2078 | 7813 |
| Surgeries | 16 | (14) | 9160 | 11,046 | 6025 | 4992 | 8521 |
| Non-surgical day cases | 21 | (19) | 2250 | 4112 | 684 | 468 | 886 |
| Outpatient visits | 92 | (81) | 1893 | 884 | 1693 | 1209 | 2262 |
| Imaging | 92 | (81) | 1336 | 971 | 1118 | 571 | 1847 |
| Total costs | 113 | (100) | 20,327 | 28,368 | 7293 | 2495 | 30,976 |
When considering descriptive statistics of cost breakdowns (note, formal statistical comparisons were not made of the breakdowns), the largest cost contributor was biologics, which manifested as a greater proportion of patients receiving these more expensive therapies: 43/81 (53%) of those patients developing disabling disease within 5 years received biologics, compared with 44/113 (39%) of patients who did not develop disabling disease. This led to a higher mean drug cost in the disabling disease group. Another difference was inpatient stays: a greater proportion of patients developing severe/disabling disease had these (53% vs. 17%), resulting in a higher mean cost per patient (£10,994 vs. £5427). Of note, however, hospitalisation is part of the calculation of the modified Beaugerie criteria so will likely have influenced this observation. Similarly, patients who developed disabling disease underwent more surgeries than those who did not (31% vs. 14%). Numbers of non-surgical day case visits, outpatient visits and imaging events appeared slightly lower in the group without disabling disease, but the differences were small.
Appendix 10, Table 33 shows unadjusted mean (SD) 5-year costs calculated separately for the nine clinical categories specified above, and for presence versus absence of severe/disabling CD (modified Beaugerie criteria).
Regression analysis
Results of separate unadjusted regressions using each of the variables listed above are given in Table 20. According to these unadjusted models, the following factors were associated with driving higher costs over the first 5 years from diagnosis: age under 40 years at diagnosis; perianal disease at diagnosis; severe (ileocolonic) endoscopic disease at diagnosis.
| n | Mean | SD | p-value | 95% CI | |
|---|---|---|---|---|---|
| sMARIA (continuous) | |||||
| Raw | 194 | 0.035 | 0.023 | 0.125 | −0.010 to 0.079 |
| Standardised | 194 | 0.010 | 0.006 | 0.125 | −0.003 to 0.022 |
| LI (continuous) | |||||
| Raw | 194 | 0.030 | 0.041 | 0.454 | −0.049 to 0.110 |
| Standardised | 194 | 0.005 | 0.007 | 0.454 | −0.008 to 0.018 |
| MEGS (continuous) | |||||
| Raw | 194 | 0.006 | 0.005 | 0.204 | −0.003 to 0.016 |
| Standardised | 194 | 0.009 | 0.007 | 0.204 | −0.005 to 0.023 |
| Beaugerie – no disabling disease at 5 years | |||||
| Beaugerie – yes disabling disease at 5 years | 194 | 0.381 | 0.197 | 0.053 | −0.005 to 0.767 |
| Age – under 40 years at diagnosis | |||||
| Age – over 40 years at diagnosis | 194 | −0.566 | 0.214 | 0.008* | −0.986 to −0.146 |
| Sex – male | |||||
| Sex – female | 194 | 0.272 | 0.194 | 0.160 | −0.107 to 0.652 |
| Smoking status – not smoker at diagnosis | |||||
| Smoking status – smoker at diagnosis | 177 | 0.210 | 0.212 | 0.320 | −0.204 to 0.625 |
| Weight loss – no | |||||
| Weight loss – recent weight loss at diagnosis | 168 | 0.248 | 0.216 | 0.249 | −0.174 to 0.671 |
| Disease behaviour – not stricturing or penetrating | |||||
| Disease behaviour – stricturing | 194 | 0.417 | 0.272 | 0.125 | −0.115 to 0.950 |
| Disease behaviour – penetrating | 194 | 0.425 | 0.286 | 0.137 | −0.135 to 0.985 |
| Perianal disease – absent | |||||
| Perianal disease – present | 193 | 0.698 | 0.313 | 0.026* | 0.085 to 1.310 |
| Severe endoscopic disease – absent | |||||
| Severe endoscopic disease – present | 177 | 0.510 | 0.240 | 0.034* | 0.039 to 0.981 |
| Location of endoscopic disease – ileocolonic | |||||
| Location of endoscopic disease – ileal | 194 | −0.674 | 0.207 | 0.001* | −1.079 to −0.269 |
| Location of endoscopic disease – colonic | 194 | −0.346 | 0.265 | 0.192 | −0.867 to 0.174 |
| Location of endoscopic disease – upper | 194 | −0.489 | 0.654 | 0.454 | −1.771 to 0.793 |
| Need for steroid therapy – no | |||||
| Need for steroid therapy – yes | 194 | 0.371 | 0.193 | 0.054 | −0.007 to 0.749 |
The best-fit regression according to the AIC and residual plot for modelling the total 5-year healthcare costs in this patient group included the modified Beaugerie criteria for severe/disabling disease within 5 years (yes or no), age at diagnosis (split into two categories: above or below 40 years of age), sex (male or female), smoker at diagnosis (yes or no), recent weight loss prior to diagnosis (< 5 kg or >= 5 kg), behaviour of disease at diagnosis (stricturing, penetrating or neither), presence of perianal disease at diagnosis, severe endoscopic disease at diagnosis, disease location at diagnosis (ileocolonic, ileal, colonic or upper digestive tract), and need for initial steroid therapy as well as two of the three MRE indices (LI and sMARIA). Removal of the MEGS improved the fit. The model used gamma family with log-link. The coefficients and other details for this model are given in Appendix 10, Table 34.
A second model with fewer covariates gave the best fit according to the BIC, although use of this summary value can be associated with under-fitting, and indeed the residuals plot was less good with this model. However, we note it here as it corroborates the statistically significant covariates identified in the best-fit regression model (see Appendix 10, Table 35).
Prevalence-based cost-of-illness calculation
The prevalence of CD is 106 per million people in the UK.71 The Office for National Statistics population estimates the UK population at mid-year 2021 to be 67.0 million.
Cost of illness for those developing disabling disease within 5 years:
Our analysis suggests 41.75% of newly diagnosed patients develop disabling disease by 5 years. Using the prevalence estimate of 106 per million patients, this equates to 106/100 * 41.75 = 44.26 people per million developing disabling disease within 5 years. This corresponds to 44.26 * 67.0 = 2965.1 people in total for the UK. The mean (SD) 5-year cost for those who have developed disabling disease is £29,763 (£38,278). So, over a 5-year period, the cost would be £29,763 * 2965.1 = £88,250,271, using the assumptions listed above.
Cost of illness for those not developing disabling disease within 5 years
Our analysis suggests 58.25% of newly diagnosed patients do not develop disabling disease within 5 years of diagnosis. Using the CD prevalence estimate of 106 per million people, this means that 106/100 * 58.25 = 61.75 * 58.25 = 61.75 people per million are CD patients who have not developed disabling disease within 5 years. This corresponds to 61.75 * 67.0 = 4137 people in total for the UK. The mean (SD) 5-year cost for those who have not developed severe/disabling disease is £20,327 (£28,368). Over a 5-year period, the total cost would therefore be £20,327 * 4137 = £84,090,766, when using the assumptions listed above.
It is important to note that performing a COI study does not provide information regarding whether resources are well spent. COI estimates are intended to illustrate which baseline patient and disease characteristics lead to higher costs as a marker for greater care needs. Future work could include this cost information in a health economic analysis that incorporates clinical outcomes and makes comparisons between different patient pathways.72