Estimating resource use to inform health technology assessments (HTA) often poses a challenge for analysts due to a lack of good quality contemporary data. Significant improvements have been made in the accessibility of routinely collected data; however, such data are not always available, complete, or presented in a level of detail required for an HTA. In the United Kingdom, National Health Service (NHS) providers collect and record data based on Healthcare Resource Groups (HRGs), designed as grouping of treatments with similar clinical characteristics and resource use (1). However, not all procedures have HRG codes, and the codes may not be sufficiently disaggregated by specific population of interest to be useful in economic evaluation.
Alternatively, resource use may be estimated from published data or primary data collection in clinical trials and observational studies. When using previously published data, simply inflating prices using inflation scale (and foreign exchange rate where applicable) is an approach often taken, but this ignores possible changes in clinical practice. Expert opinion is often used to fill in data gaps and/or supplement the trial and observational data (Reference Sullivan and Payne2). In an expert elicitation study, experts are asked to formulate a quantitative judgment based on their own beliefs (independently of the quality of such knowledge) for an uncertain quantity by augmenting existing knowledge or by formulating a probabilistic judgement over the quantity (Reference O'Hagan, Buck and Daneshkhah3;Reference Soares, Bojke and Dumville4). Undertaking such exercises requires time and resource, and the added benefit of this compared with a simple inflation adjustment is not currently clear.
Using an example of liver transplantation, we compared alternative methods for estimating contemporary resource use costs. In the European Region, there are an estimated 15 million people with hepatitis C virus (HCV), 2.0 percent of adults and a significant number of those who are chronically infected will develop liver cirrhosis or liver cancer and may require transplantation (1). Liver transplantation is an effective treatment option for end-stage liver disease and acute liver failure and has been available in the United Kingdom NHS since 1983. In the United Kingdom, alcohol and chronic HCV are the leading causes of liver transplantation (Reference Sullivan and Payne2;Reference O'Hagan, Buck and Daneshkhah3). A total of 775 liver transplants were carried out in the NHS in 2013 (Reference Soares, Bojke and Dumville4).
Contemporary data on the costs of treating patients undergoing liver transplantation are lacking. Recent HTAs of treatments for HCV and liver disease have frequently used data from the Cost-Effectiveness of Liver Transplant (CELT) study (Reference Grieve, Roberts and Wright5–Reference Crossan, Tsochatzis and Longworth8). CELT was a large cohort study commissioned by the Department of Health to assess all patients considered for liver transplantation in England and Wales in 1995–96, and included the collection of detailed healthcare resource-use information for all patients (Reference Longworth, Young and Buxton9). Researchers using these data have applied simple inflation adjustments to reflect changes in costs in the 20 years since the study was conducted (Reference Grieve, Roberts and Wright5–Reference Crossan, Tsochatzis and Longworth8); however, this will be inaccurate if changes in significant clinical practice have occurred in the interim. If efficiencies were made in delivery of care over that period, simple inflation of total cost or updating unit prices only would overestimate the cost of care. Systematic bias would enter any economic analysis based on it and subsequent policy decision.
This study aims to compare alternative approaches to updating resource cost estimates using liver transplantation for patients with chronic HCV as an example.
METHODS
The potential impact on resource costs and savings of a health technology or procedure is estimated in HTAs by first ascertaining resource use: this is done by identifying the resource item first and then measuring it in terms of frequency (e.g., number of visits to general practitioner), intensity (e.g., daily dosage of medication) and duration of the healthcare activity (e.g., length of stay in the general ward) used by the patient. The total amount of resource use is then multiplied by corresponding unit cost to generate the total resource cost for each item. National estimates of unit prices, for example NHS Reference Costs, are applied to resource use amount to increase generalizability.
Resource use data were identified from the CELT study and updated using four different methods: (i) simple inflation; (ii) updating unit cost estimates; (iii) updating key resource use estimates based on expert opinion and updating unit costs; (iv) updating key resource use estimates based on registry data and updating unit costs. Please note that the first two methods implicitly assume that resource use amongst liver transplant patients has not changed since CELT was conducted over 15 years ago.
CELT included 755 patients assessed for a liver transplant from the six transplant units in England; 477 of whom received a transplant. Detailed resource use data were collected from the assessment of patients for transplant suitability to 2 years posttransplantation on discharge from transplant centers (Reference Longworth, Young and Buxton9). Most unit costs were obtained from the transplant centers and based on 1998/9 £ GBP prices. Drug costs were sourced from the 1999 edition of the British National Formulary (BNF) (10). Of the patients who were assessed in the CELT study, ninety-eight had a primary diagnosis of HCV. A total of sixty-seven patients with HCV underwent liver transplantation, of which sixty-six were followed up in the 2 years posttransplant. This study has been reported in greater detail elsewhere (Reference Longworth, Young and Buxton9).
Resource use data were collected in the CELT study over four phases reflecting key stages in the transplantation procedure. The assessment phase started from date of admission for assessment of suitability of liver transplantation to date of listing for transplantation. For those patients who were not listed for transplantation, the date of discharge was used as the end date. The candidacy phase started from the date of being added to the transplant waiting list to the date the patient was admitted for their transplant operation. The transplant phase was from admission for the transplant operation to discharge following the operation. Finally, the posttransplant phase was defined as date of discharge following the operation for 2 years. The main cost drivers in the CELT study for patients with HCV were length of inpatient stay and duration of transplant operation. In addition, cost of outpatient visits, dietician and physiotherapy sessions, and some key drugs (e.g., immunosuppressant), tests, and treatments also impacted upon the total cost of liver transplantation.
Method 1: Simple Inflation
For simple inflation, the total cost of liver transplantation was updated using inflation indices. An intention to treat approach was taken for completeness. All patients assessed for liver transplantation were included in the cost study regardless of whether they made it to the subsequent phases of candidacy, transplant operation, and follow-up. The mean total cost of liver transplantation for HCV patients in the CELT study was £43,283 and the value ranged from £405 to £302,470 in 1998 prices. The low estimate at the bottom of the range reflects the costs of a patient who was placed on the transplant waiting list but did not receive a transplant because they were too ill to be operated on or died pretransplantation. The cost was updated using a multiplier calculated from the pay and price indices for Hospital & Community Health Services (HCHS) for the years 1998 and 2013 (Reference Crossan, Tsochatzis and Longworth8).
Method 2: Updating Unit Cost Estimates
This approach comprised extraction of resource use data from the CELT study and applying unit cost data from routinely available national sources. Unit costs of inpatient stay were taken from NHS Reference Costs 2012–13 (1). The prices and recommended doses of immunosuppressant drugs were sourced from the BNF (11). The unit costs of key tests and treatments were sourced from NHS Reference Costs 2012–13 where available, except for biopsy, which was sourced from a recent HTA report from the United Kingdom (1;Reference Stevenson, Lloyd-Jones, Morgan and Wong12). The costs of remaining resource use items were inflated using a HCHS multiplier (13). The unit cost data were applied to resource use data to estimate total cost for HCV patients.
Method 3: Expert Elicitation and Updated Costs
The third approach used expert elicitation to ascertain the key changes in clinical practice of HCV patients undergoing liver transplantation since the mid-1990s and to obtain estimates of some key items of resource use. Expert elicitation comprised of face-to-face interviews with clinical experts in liver transplantation. The aim was to obtain views from professionals with differing expertise from a range of transplant centers and to recruit professionals involved directly in liver transplantation. Brunel Research Ethics Committee provided ethical clearance to conduct the expert elicitation study. Invitation letters and information sheets were sent to consultant hepatologists, surgeons, and transplant co-ordinators identified from seven transplant centers in the United Kingdom. Invitations were sent by a combination of post and email.
The first part of the interview took a semi-structured format and included open-ended questions about how the clinical management of patients with HCV undergoing liver transplantation has changed since mid-1990s. The questions were related to changes in four main areas: clinical practice, patient case-mix, outcomes, and organ retrieval. The experts were also asked questions regarding changes in specific tests and treatments in the clinical management of chronic HCV patients undergoing liver transplantation. A short list of high cost/high volume tests and treatments were identified from the CELT dataset and presented to the experts for their views on their current use. The procedures that were chosen from CELT were: paracentesis, ascitic tape, endoscopic retrograde cholangiopancreatography, endoscopy, continuous veno-venous hemofiltration (CVVH), laparotomy, angiography-hepatic, echocardiogram, multigated acquisition scan (MUGA), magnetic resonance imaging of the abdomen, and electroencephalogram. Additional questions were asked relating to practices that have been discontinued in the past 15 years and new ones that have been introduced.
In the second part of the interview, the experts were asked to estimate some specific healthcare resources used in the management of these patients based on key drivers of costs identified from the CELT study. The questions focused on length of operation and inpatient stay in ward or intensive treatment unit (ITU) by phase, across three phases: pretransplant, transplant, and posttransplant. The assessment and candidacy phases were combined into a pretransplant phase as distinguishing between these was found to be difficult during the CELT study. The posttransplantation phase was further split into first year posttransplant and subsequent years.
An expert elicitation tool adapted from Leal at al. was used to elicit these resource use estimates (Reference Leal, Wordsworth, Legood and Blair14). The tool allowed for variability around the resource use estimates to be captured using probability distributions. The main reason being that focusing only on the “typical” or “average” patient may misrepresent resource use and some patients require more resource-intensive treatment, while others require less (Reference Stevenson, Lloyd-Jones, Morgan and Wong12;13). Resource use estimates were sought for a “typical” or average patient with HCV (M), followed by lower (L) and higher (H) estimates to represent the variance around the resource use estimate. Four complementary intervals were calculated using the estimates M, L, and H as shown in Eq. [1].
$$\begin{equation}
\left[ {L,\frac{{L + M}}{2}} \right];\left[ {\frac{{L + M}}{2},M} \right];\left[ {M,\frac{{M + H}}{2}} \right];\left[ {\frac{{M + H}}{2},H} \right].
\end{equation}$$
The experts were asked to approximate the proportion of patients likely to receive the amount of care for each interval and a histogram displaying estimates and probabilities provided by the expert was produced in real time. The histogram was shown back to the expert to check whether it represented his or her beliefs. An interactive user-interface was developed in Microsoft Excel 2010 for this exercise, an example screen of which is illustrated in Figure 1. If the histogram failed in representing the expert's opinion, an option was given to repeat the exercise. Depending on their clinical specialty and role in each phase of transplant care, some respondents found it difficult to respond to questions about the frequency of specific procedures and were allowed to omit specific questions if outside their expertise.
Figure 1. An example of expert elicitation tool used.
A summary of the elicited estimates was compiled and shared with the participating experts by email. The probability distributions elicited from experts were summarized using the vincentization method reported by Leal et al. (Reference Leal, Wordsworth, Legood and Blair14). A simple linear opinion pool method was used to aggregate the results across 0–1 interval space. The probability density function (PDF) for each expert was allocated over small dimensions of that interval space. The PDFs of the experts for each interval was summed and was divided by the number of experts to get an aggregated probability distribution. The qualitative data were analyzed using a framework analysis method reflecting the semi-structured nature of the interview. New unit costs obtained from national sources were applied to expert elicitation data to re-estimate cost of transplantation.
Method 4: Use of Routine Data and Inflation
The fourth approach to estimate costs included observational data on resource-use from the NHS. There are limited observational data in the public domain regarding resources use in liver transplantation in the United Kingdom. NHS Blood and Transplant (NHSBT) is a Special Health Authority to promote and manage blood, tissue, and organ donation and routinely collects resource use information on liver transplantation. We contacted NHSBT for access to routine use data, and in May 2014 they provided us with observational data for all hepatitis patients, HCV and hepatitis B (HBV), who received liver transplantation in the United Kingdom between 2008 and 2013. It included transplant details such as date of transplant, cold ischemia time, veno-venous bypass time, operative reperfusion time (liver out of ice to reperfusion with blood), inpatient stay in transplantation ITU (ward, ITU, and ventilation), and posttransplantation inpatient stay (ward and ITU). Information on pretransplantation was not available from NHSBT. Current unit costs were applied to new resource use data to re-estimate cost of transplantation.
RESULTS
In this section, we present the cost estimates and resource use identified using the four approaches followed by our re-estimation of total cost of liver transplantation and comparison of methods.
Simple Inflation
The total cost of liver transplantation for patients with HCV simply updated to current prices was £87,432. However, this method does not take into account breakdown of cost by resource use of phase of transplantation.
Updating Unit Cost Estimates
Most unit costs had increased since the CELT study. A summary of updated unit costs are in Tables A and B of the online Supplementary File. Key exceptions were the unit costs of immunosuppressant therapy, which have decreased since the CELT study was conducted. The new total cost of transplantation using this method was £116,009.
Expert Elicitation
Letters of invitation were initially sent to fourteen experts by means of post in mid-December 2013. A total of three or four contact attempts were made to recruit the experts following the initial letter. To increase the sample size, a further twenty-one clinical practitioners were contacted. In total, six full interviews were conducted. An additional expert was identified during one of the interviews and responded to the open-ended question section of the survey, but it was not possible to complete the quantitative elicitation exercise. The interviews took place between December 2013 and early March 2014, and comprised of two hepatologists, three surgeons, and one transplant co-ordinator from four different liver transplant units in the United Kingdom. The six interviews were recorded and transcribed and lasted 30–60 minutes.
The recurrent themes in each topic area were identified and summarized according to three different phases: pretransplant, transplant, and posttransplant. All the participants in the study stated that clinical practice differed between the liver transplant units and treatment protocols varied. In addition, it was stated that the number of patients treated, severity of liver disease, and number of comorbidities varied across transplant units. After the six interviews, few new themes were raised by the experts, indicating that saturation was achieved.
Some participants noted that the number of patients on the waiting list for liver transplant has increased and the reason being twofold: an increase in number of people being referred for transplant and the supply of donor organs not matching demand. They stated that, while the number of organs from donors after cardiac death (DCD) had increased rapidly, organs from donors after brain death (DBD) had only marginally increased, if at all. Few experts also commented that recurrence of HCV is very common (compared with transplantation due to other indications) and may have increased due to the use of sub-optimal organs. Some of the experts reported an increase in referral to other providers for monitoring, such as local care providers.
Regarding tests and procedures carried out on liver transplant patients, more than one expert said that use of angiography and MUGA have decreased with increased use of echocardiogram, laparotomy and CVVH, and surgeons mentioned that bone-scan procedures during the transplant operation have discontinued. One expert referred to new clinical practices and support that are routinely provided and were introduced since the CELT study, such as cardio-pulmonary exercise testing, detailed cardiovascular disease scanning, and rifaxamin to treat hepatic encephalopathy. Two specific changes in standard drug therapy were highlighted by several experts: tacrolimus replacing cyclosporine as the immunosuppressant of choice, and azathioprine being substituted by mycophenolate mofetil (MMF) for maintenance treatment.
The second part of the interview focused on eliciting resource use estimates from the experts, specifically inpatient stay and operation duration. The variation in resource use data among patients was very large and reflected differences in comorbidities and prognosis of patients. The combined probability distribution of resource use parameters are presented in Figure 2.
Figure 2. Summary of resource use estimates from expert elicitation.
Registry Data
Observational data from routine care of patients admitted for liver transplantation were provided by NHSBT in May 2014 for all patients with HCV and HBV who received liver transplantation in the United Kingdom over a 5-year period. The mean age of liver transplant recipients was 52 years, which is comparable to the national profile of all liver transplant recipients as well as CELT study. One-third of patients had had more than one liver graft. Almost all patients received whole liver transplants apart from a case each of reduced and split liver transplant. A little over one-third of the HCV and HBV patients who were transplanted died in the 5-year period.
DBD liver was used in almost 90 percent of the patients transplanted between 2008 and 2013, and only three patients received a DCD liver. In the CELT study, all patients only received DBD liver, with other types of liver transplant introduced in the NHS only recently. A comparison of resource use estimates from the CELT study only, expert elicitation study, and NHSBT dataset is presented in Table 1.
Table 1. Resource Use Data from Different Sources
HCV, hepatitis C virus; HBV, hepatitis B virus; SD, standard deviation; ITU, intensive treatment unit; CELT study, Cost-Effectiveness of Liver Transplant study; NHSBT, National Health Service Blood and Transplant.
Re-estimating Total Cost of Liver Transplantation
The updated total costs of liver transplantation using different data sources are presented in Table 2 for comparison. Two key changes were made: the length of inpatient stay from CELT study was replaced with mean estimate aggregated from the expert elicitation and NHSBT. And the changes to main drug treatments were incorporated to reflect tacrolimus being replaced by cyclosporine as the standard immunosuppressant for patients (assumed 95 percent tacrolimus and 5 percent cyclosporine based on the expert elicitation). The costs were also adjusted to reflect azathioprine being substituted by MMF for maintenance treatment.
Table 2. Updated Costs of Treating Patients with HCV
HCV, hepatitis C virus; HBV, hepatitis B virus; SD, standard deviation; NHSBT, National Health Service Blood and Transplant.
The NHSBT resource use estimates were not available for the pretransplantation phase. The resource use data from NHSBT for transplant and posttransplant were for both HCV and HBV patients. The mean total cost of liver transplantation, excluding pretransplantation, was estimated at £89,598, £93,510, and £103,156 per patient using updating unit prices, expert elicitation, and NHSBT information on resource use. Including pretransplantation the total cost was £87,321, £116,009, and £111,565 using simple inflation, updating unit cost estimates, and expert elicitation, respectively. The NHSBT data are the most accurate reflection of cost estimates for transplantation and posttransplantation phase; however, given lack of NHSBT data for pretransplant phase and given the changes in clinical practice, expert elicitation for pretransplant was added to give a total cost of £121,211.
Comparison of Methods
The first and second methods did not account for changes in clinical practice and yielded similar cost estimates for the pretransplantation phase; however, the results were different for transplantation and follow-up. The changes in unit cost of the key items were mixed, for example, the cost of immunosuppressant drugs had reduced but the cost of tests had increased over time. The primary reason in difference is the sample used. The first method focused on patients with primary biliary cirrhosis, alcoholic liver disease, and primary sclerosing cholangitis, while the second method is specific to hepatitis patients.
The use of experts to confirm current clinical practice was very useful; however, the quantification of resource use using experts is less clear. The experts suggested that there have been reductions in length of stay for patients in pretransplantation phase. This is likely to be due to more streamlined assessment processes, but is also likely to be in part due to differences in methods of estimation between the CELT study and the elicitation study. The elicitation exercise focused on the resources associated with preparation for transplantation only whereas the CELT study also captured some resource use corresponding to the ongoing treatment of end-stage liver disease and associated complications. Second, in the transplantation phase expert elicitation assumed all patients had transplantation, while the CELT study captured intention to treat patient group and perhaps other complications as well based on the range captured.
The posttransplantation cost estimated from the NHSBT data is much higher than the cost estimated using other methods. The reason could be that patients are often treated by local healthcare providers after discharge, which may be captured by NHSBT but not by the expert elicitation and CELT focused on resource use within the transplant units.
NHSBT data are the preferred dataset, which reflect current practice and capture resources incurred by all NHS providers, but they do not provide a breakdown on resource use for all four phases. Liver transplantation is, however, an exception and central registry data are not available in most disease areas.
DISCUSSION
Our study has provided current estimates of liver transplantation cost to transplant centers in England, in total and for three key phases of treatment: pretransplantation, transplantation, and posttransplantation. The study compared various methods to update cost estimate. The mean total cost of liver transplantation was estimated at £87,432, £116,009, and £111,565 per patient using simple inflation, updating unit cost estimates, and expert elicitation, and total cost of £121,211 was estimated when combining expert elicitation and NHSBT. Nevertheless, the variation in cost is less than 10 percent when using the four approaches.
In the first method, we simply inflated total resource cost using an estimate from the literature and applying inflation over time. The method is quick and easy but does not contain a breakdown of cost by actual disease or resource use, and does not take into account any changes in practice. In the second approach, the resource use data for HCV and HBV patients were extracted from a robust observational study from over 15 years ago was used and only unit costs were inflated to reflect change in prices. This method of simply inflating unit prices is inaccurate when there have been changes in clinical practice as was identified from the expert elicitation exercise. There have been some significant changes in how HCV patients are treated since the original cohort study was conducted 15 years ago. Experts suggested decreased resource use in the pretransplantation phase compared with CELT because of more streamlined assessment processes in place and some of the assessments being done outside the unit. Regarding the transplant phase, costs calculated using simple inflation of CELT and NHSBT were similar but experts indicated slightly higher estimates.
As our third method, we used expert elicitation to identify key changes in clinical practice and estimated key resource use items, before applying inflated unit costs. Expert elicitation is “appropriately viewed as being second best and a potential source of parameter un-certainty” (Reference Sullivan and Payne2). The main reasons cited are that the experts have limited experience and perspective and that the methods used to identify the experts and elicit their opinion is not robust. However, in this study, several steps were taken to obtain the best estimates possible from expert elicitation exercise. All of the experts were highly experienced in the treatment of patients with HCV and liver transplantation. As each expert may be involved at different stages of patients’ care, they were selected to reflect different perspectives from range of clinical professions and a range of transplant centers in the United Kingdom.
The experts indicated a higher mean estimate of days in ward and ITU. They argued that patients with greater number of comorbidities are now being considered for transplantation and to meet increase in demand, sub-optimal (DCD) organs are being used more frequently, both of which have resulted in increased inpatient stay. In the posttransplant phase, the resource use suggested by experts and CELT was similar; however, NHSBT data revealed much higher resource use following discharge after transplant operation. The latter is likely to be correct, because some experts had admitted to not knowing about follow-up as it is usually provided outside the transplant unit and also stated recurrence of HCV as common.
Finally, registry data of patients undergoing liver transplantation were obtained and inflated unit costs were applied to re-estimate the cost of liver transplantation. Observation data of routine care collected by NHSBT are potentially the most reliable reflection of current resource use in the United Kingdom. However, such data are rarely available and accessible. Another potential limitation is that information covering entire patient pathway may not be available; for example, it was not possible to estimate total cost of liver transplantation using NHSBT data as it did not have any data on resource use before transplantation, or it is not aggregated by disease group of interest. Our preferred estimate is using NHSBT estimates for liver transplantation combined with expert elicitation for pretransplantation phases, which gives a total cost of £121,211.
It should be noted that it is rare to have registry data so readily available. Routine collection of resource use data is costly and poses an administrative burden; nevertheless, it is very useful and should be encouraged in other disease areas as well. Also use of electronic record keeping of resource use of patients will perhaps improve transfer of information to those working on HTA.
Liver transplantation remains a costly health intervention for patients with HCV. Also note that our estimates do not include the costs of harvesting donor organs as it was not possible to identify a robust estimate from experts or through literature review.
CONCLUSION
We have estimated the resource use to NHS transplant units of liver transplantation for patients with chronic HCV, using high-quality historical data, expert elicitation, and observational data on routine care where available. Each approach had its limitation. Simple inflation and updating unit cost estimates of resource use is a quick way of updating cost data, but one has to be mindful of changes in clinical practice when using historical data. Using routine data is preferable; however, it may not be available for all items of resource use or in the level of detail or format required. When no routine data are available and simply inflating historical data is inadequate, expert elicitation (conducted robustly) provides a reasonable estimate of resource use.
The mean total cost of liver transplantation was estimated at £87,432, £116,009, and £111,565 per patient using simple inflation, updating unit cost estimates, and expert elicitation. In the case of liver transplantation, the simple inflation method underestimated total costs. There have been significant changes in how HCV patients are treated since the original cohort study was conducted 15 years ago, and this is reflected in the breakdown of costs at different phases (pretransplant, operation, and posttransplant). Our preferred estimate is using NHSBT estimates with expert elicitation for pretransplantation, which gives a total cost of £121,211.
SUPPLEMENTARY MATERIAL
Supplementary Table 1: https://doi.org/10.1017/S026646231700071X
Supplementary Table 2: https://doi.org/10.1017/S026646231700071X
CONFLICTS OF INTEREST
The authors have nothing to disclose.
