|
Economic evaluation of insulin lispro versus neutral (regular) insulin therapy using a willingness to pay approach |
Davey P, Grainger D, MacMillan J, Rajan N, Aristides M, Dobson M |
|
|
Record Status This is a critical abstract of an economic evaluation that meets the criteria for inclusion on NHS EED. Each abstract contains a brief summary of the methods, the results and conclusions followed by a detailed critical assessment on the reliability of the study and the conclusions drawn. Health technology The study investigated the use of insulin lispro, a new insulin at the time of study, in the treatment of diabetes mellitus. Insulin lispro (Humalog(R)), an analogue of human insulin, is a new biosynthetic insulin analogue created by reversing the amino acids at positions 28 and 29 on the insulin B chain. This alteration results in an insulin with a faster rate of absorption from subcutaneous injection sites, a more rapid onset of action, and a shorter duration of action than human regular insulin. Insulin lispro has its onset of action within 15 minutes of administration, compared with 30 to 45 minutes for human regular insulin. Peak concentrations are higher and occur earlier than with human regular insulin. This more rapid onset allows injection at, or close to, mealtimes.
Study population Patients between 17 and 70 years of age, diagnosed with type I or II diabetes mellitus and who had used either insulin lispro or neutral insulin over the previous 2 months.
Setting The study setting was hospital. The economic study was carried out in Australia.
Dates to which data relate Effectiveness and resource use data were obtained from Eli Lilly and Company in 1994. The time frame during which the willingness to pay (WTP) study was carried out was not specified. The price year was not given.
Source of effectiveness data Evidence for the final outcome was derived from a review of previously completed studies and an expert panel.
Outcomes assessed in the review The outcome measure used in the meta-analysis was the proportion of patients achieving at least 1 therapeutic success from the following:
a post-prandial blood glucose level of 8mmol/L or less;
a 2-hour post-prandial blood glucose level within 20% of the pre-meal level; or
a decrease from baseline in 2-hour post-prandial blood glucose of at least 50%.
The data were analysed for types I and II diabetes, both separately and combined. P-values and odds ratios were reported extensively in a separate paper containing details of the meta-analysis (see Other Publications below).
A separate meta-analysis of two pivotal studies in type I patients was performed to provide data relating the frequency of hypoglycaemia to the level of overall glycaemic control.
Study designs and other criteria for inclusion in the review The a priori inclusion criteria for the meta-analysis were that studies should be randomised, controlled trials of insulin lispro versus human regular insulin, be of at least 6 months duration, and enroll more than 30 patients with type I (insulin-dependent) or type II (non-insulin-dependent) diabetes who had previously been managed on insulin therapy.
It was noted that the patients with type II diabetes and who were new to insulin therapy had a wide range of disease severity (newly diagnosed, currently on oral therapy, and oral therapy failures). For these reasons, it was felt that these studies might not be representative of the use of insulin lispro in the general community and might bias the results of the meta-analysis. Subsequently, the authors also conducted a meta-analysis to include these studies, with no significant change in the results. The studies eventually included in the review were phase III, open label, randomised controlled trials undertaken to support the New Drug Application (NDA) for insulin lispro.
Sources searched to identify primary studies Two approaches were taken to identify all trials comparing insulin lispro with human regular insulin. Firstly, reports of all phase III trials conducted by Eli Lilly and Company (Indianapolis, Indiana) were obtained from the New Drug Application age package. The second approach to ensuring consideration of all available studies involved conducting a literature search according to published guidelines. A MEDLINE CD-ROM search using SilverPlatter was conducted. A dial-up MEDLINE search was also conducted (1966 to 1996), as well as an EMBASE dial-up search (1974 to l996).
Criteria used to ensure the validity of primary studies The authors assessed the trials according to the criteria in the Guidelines for the Pharmaceutical Industry on Preparation of Submissions to the Pharmaceutical Benefits Advisory Committee. All the studies were randomised and followed both parallel and crossover designs. It was considered appropriate to pool parallel-design and crossover studies as long as no carryover effect interfered with outcomes in the second phase of the trial.
Assessment of carryover indicated that only one study showed evidence of any carryover effect in the area of 2-hour blood glucose excursion, which is a measure of the way in which blood glucose levels rise after a meal and drop with the post-prandial excretion (in normal individuals) or injection (in diabetic individuals) of insulin. This carryover effect appeared to be attributable to changes in the basal insulin dosage in the second phase of the study and had no impact on the results of the meta-analysis. The studies were all open-label due to the objective of maintaining optimal injection timing relative to meals. Failure to meet this objective (for example using identical injection timing relative to meals) would have resulted in one or the other of the study insulins being used in a suboptimal manner, thus biasing the results. All study results were reported on an intention to treat basis.
Methods used to judge relevance and validity, and for extracting data Number of primary studies included 6 trials out of 8 were included in the meta-analysis: two other trials were excluded because the patients were new to insulin therapy. Three studies identified by the literature search were not included in the meta-analysis since each focused on measurements of short-term pharmacokinetic and glucose response.
Methods of combining primary studies A meta-analysis was conducted to combine the primary studies; patients with type I and type II diabetes and all patients combined being considered. The pooling of type I and type II patients was considered useful to provide an overall summary measure of relative efficacy in the diabetic population. The relative efficacy of insulin lispro was found to be similar in both patient populations, which suggests that pooling trials in both types of patients was appropriate. Trials in type I and II patients were also pooled separately to demonstrate any differences in relative efficacy in these subgroups.
Investigation of differences between primary studies It was reported that all of the trials were conducted as part of a coordinated clinical trial programme, and there was considerable uniformity across the studies in all aspects of trial design. The chi-square statistic was used to test for lack of homogeneity.
Results of the review The meta-analysis demonstrated statistically significant differences in 2 outcomes: patients achieving at least 1 therapeutic success and patients gaining a post-prandial blood glucose of 8 mmol/L or less. The proportion of patients gaining at least 1 therapeutic success for both types of diabetes (type I and type II) was 59.4% for the insulin lispro group and 49.3% for the neutral insulin group. The proportion of patients gaining a post-prandial blood glucose less than or equal to 8 mmol/L was 25.5% and 17.6%, respectively.
The proportion of patients gaining at least 1 therapeutic success for type I patients was 62.5% for the insulin lispro group and 51.7%% for the neutral insulin group. The proportion of patients gaining a post-prandial blood glucose level of 8 mmol/L or less was 29.7% for the insulin lispro group and 21.7% for the neutral insulin group.
The proportion of patients gaining at least 1 therapeutic success for type II patients was 55.3% for the insulin lispro group and 46% for the neutral insulin group. The proportion of patients gaining a post-prandial blood glucose of 8 mmol/L or less was 19.8% for the insulin lispro group and 12% for the neutral insulin group.
The odds of patients with type I and type II diabetes achieving at least one therapeutic success or of achieving a post-prandial blood glucose level of 8 mmol/L or less were significantly higher with lispro than with human regular insulin, (p<0.00001).
The meta-analysis of glycosylated haemoglobin levels (HbAlc) versus frequency of hypoglycaemic events in 336 patients (from two pivotal studies) with type I diabetes mellitus showed that patients receiving neutral insulin had an increased rate of hypoglycaemia with decreasing levels of HbAlc, (p=0.006). Patients on insulin lispro had no significant increase in hypoglycaemia. The clinical importance of this finding is that lower HbAlc levels have been demonstrated to reduce the microvascular complications of diabetes.
Methods used to derive estimates of effectiveness An expert panel of 3 endocrinologists and a diabetes nurse specialist was established to evaluate the clinical significance of the statistically significant results of the meta-analysis.
Estimates of effectiveness and key assumptions The meta-analysis results of two outcomes with statistical significance (as reported above) were deemed to be clinically significant by the expert panel.
Measure of benefits used in the economic analysis The benefits of the health technologies studied were expressed in monetary values: the average incremental benefit per patient per month and per year. This was calculated using the contingent-valuation (CV) approach to derive willingness to pay (WTP). The WTP study incorporated scenarios describing the outcomes from insulin lispro and neutral insulin (in a blinded fashion), the results from a formal meta-analysis and a description of the injection characteristics of the therapies. The patient's initial wi1lingness to pay was elicited by an open-ended question to eliminate starting point and range bias. Thereafter, a "bid-up" process was used to obtain the maximum willingness to pay. To increase the "realism" of the scenario, the question was asked in terms of an additional out-of-pocket expense similar to the pharmaceutical copayments that patients make in Australia.
The WTP study was designed to detect a treatment difference at alpha=0.05 and a power of 85%. The minimum mean difference in willingness to pay it was designed to detect was Aus$20. On the basis of a pilot study of 11 patients, it was estimated that a samp1e of 80 patients was required. A sample of 83 patients with type I or II diabetes mellitus were surveyed (via face-to-face interviews) in 2 large diabetes centres in 2 different cities using an open questionnaire to determine their maximum willingness to pay for the therapy they preferred.
41% of patients had participated in clinical trials of insulin lispro. The majority of the sample (87%) was happy or very happy with their current therapy and 95% indicated that they would try a new therapy if it offered advantages. Regression analysis was used to explore the relationship of willingness to pay with patient demographics. The patients' willingness to pay was analysed by a variety of demographic and clinical characteristics.
Direct costs Costs were not discounted because of the one-year time frame of the cost analysis. Quantities were based on the controlled studies. The incremental cost of the intervention drug itself was used in the cost analysis as there were deemed to be no additional costs involved with the use of insulin lispro. Insulin lispro was listed on the Australian National Formulary at a 36% premium over neutral insulin. The price year was not reported.
Indirect Costs Indirect costs were not considered.
Currency Australian dollars (Aus$).
Sensitivity analysis A single one-way sensitivity analysis was carried out based on the upper and lower bounds of the 95% confidence interval of the WTP values.
Estimated benefits used in the economic analysis 76 patients (92%) preferred insulin lispro, and 7 patients (8%) preferred neutral insulin, (p<=0.001). Once patients had selected the preferred insulin therapy, the patient's willingness to pay was similar. The WTP study indicated that the average incremental benefit per patient per month with insulin lispro was Aus$37.68 (p<0.0001, 95% CI: Aus$47.84 - Aus$27.52). 83 patients valued the additional benefit of insulin lispro at Aus$3,262.68 and neutral insulin at Aus$135.52. This represents average incremental values per patient for insulin lispro of Aus$39.31 and for neutral insulin Aus$1.63 (insulin lispro Aus$39.31 - neutral insulin Aus$1.63 = Aus$37.68). The incremental benefit of insulin lispro over neutral insulin per patient per year was Aus$452.16 (95% CI: Aus$574.08 - Aus$330.24). The subgroup analysis indicated that willingness to pay was significantly associated with age (p<0.01, exponential when centre adjusted). Patients who were middle-aged displayed the strongest willingness to pay for insulin lispro. When adjusted for age (and income) only history of hypoglycaemia approached statistical significance (p=0.06).
Cost results The annual cost of treating a patient with neutral insulin was Aus$197.88 and with insulin lispro was Aus$268.20, leading to an incremental cost per year of Aus$70.32.
Synthesis of costs and benefits Costs and benefits were combined using a net benefit approach (incremental WTP benefits minus incremental costs) in the framework of a cost-benefit analysis. The net benefit of neutral insulin over insulin lispro was Aus$381.84. Benefits (Aus$452.16) were greater than costs (Aus$70.32) under the insulin lispro. The sensitivity analysis showed that the cost-benefit results were robust. The relationship between willingness to pay and various demographic factors supported the robustness of the results.
Authors' conclusions This study showed that costs were exceeded by the benefits, and insulin lispro was deemed to offer a net benefit. A multivariate analysis indicated that those patients who were middle-aged had the strongest preference for insulin lispro.
CRD COMMENTARY - Selection of comparators A justification was given for the choice of the comparator. Neutral insulin therapy was considered to be the comparator since it was the regular treatment of choice in the context in question. You, as a database user, should consider whether this is a widely used health technology in you own setting.
Validity of estimate of measure of effectiveness Given the systematic nature of the literature review carried out by the authors, the internal validity of the effectiveness is likely to be high. The validity of the effectiveness estimates is further enhanced by the inclusion of randomised trials, the investigation of the quality of the primary studies, the evaluation of differences among studies, and the intention to treat basis for the effectiveness analysis. Furthermore, the authors demonstrated that the inclusion of the two excluded trials in the meta-analysis would not change the main set of results. The controversial issue of whether both phases of crossover trials should be included in the meta-analysis was also addressed by the authors, which served to establish the robustness of the effectiveness results.
Validity of estimate of measure of benefit Despite all the difficulties and challenges involved, the authors managed to successfully conduct the WTP study. This was the first study of its kind undertaken in Australia to support an application for listing of a new drug on the Australian National Formulary. The estimate of the WTP was based on a separate study, and not directly on the effectiveness analysis. The choice of the benefit measure was justified, as it was deemed that the WTP measurement in the framework of a cost-benefit analysis:
offered an absolute "decision-rule" for funding;
offered a means of including multidimensional outcomes (as it allows all important outcomes to be captured in a single ratio);
is useful in assisting the value-for-money judgement as it can transform narrowly used intermediate markers into more commonly applied measure of value such as monetary value; and
not only has the advantage of being able to capture externalities but also allows patients to consider the opportunity costs of healthcare expenditures in the context of non-healthcare services.
It was noted that, as a result of the use of an open-ended question, the risk of starting-point bias was eliminated. However, the WTP study suffered from lack of adequate power, as required by the prospective power analysis.
Validity of estimate of measure of costs:
The authors did not draw a clear picture of the resource use profile and cost profile of each treatment. As a result insufficient details were provided of the costing structure for the alternatives studied, and the reader was referred to another paper. Furthermore, the authors chose not to consider indirect costs (production loss) and, given the lack of full details of the resource use structure, it is hard to judge whether or not this was appropriate. Lack of information on the price year adopted in the study may hinder the generalisability of the cost results. However, the authors did report the perspective adopted in the cost analysis.
Other issues Since the study was used to support an application to the Australian National Formulary, there was no discussion of the generalisability of the result to other settings or countries.
The authors mentioned some limitations of the WTP study.
The literature suggests that studies should test for scope to ensure validity of the method. This involves varying the size of health gain in different subsamples and then testing whether willingness to pay increases with the size of the health gain. This study did not employ the method suggested in the literature to investigate the scope effect, as it would have necessitated a substantially larger sample size. However, this study did find a difference in willingness to pay between patients with type I and II diabetes mellitus, with type I patients gaining a greater clinical benefit and also recording a higher willingness to pay. This finding indicates that the study was associated with a scope effect.
This study employed an open-ended rather than a closed-question method. The open-ended method has been criticised by some authors who claim that this method is unreliable and has validity problems. In addition, in a study, patients were found to have problems responding to an open question. Given that the study used a mail-out approach the implications for this study, which used interview-administered questionnaires, are unclear.
It is likely that the chronic nature of the condition meant that patients were better informed and so were able to make informed responses to the questions posed. It is possible that the results may have been different if the patients had had to bid on a more abstract scenario that implied no direct use by the respondent.
One of the important unanswered questions with contingent-valuation methods is whether patients would actually pay the amount given in the interview. In this study, the scenarios were made as realistic as possible to encourage patients to answer accurately and in accordance with their opportunity cost.
Implications of the study In relation to the effectiveness analysis, it was suggested that longer trials would be required to demonstrate the clinical implications of the advantages offered by insulin lispro therapy.
With the WTP study the authors suggest that it is possible that there is a correlation between age and income and education (i.e. that income and education may be associated with willingness to pay). However, this would require further investigation. The implication of this finding is that the age of patients receiving insulin is the strongest determinant of their preference for insulin lispro therapy. Middle-aged patients had a stronger preference for insulin lispro relative to younger and older patients.
This study highlights the need for care in constructing the description of a drug product or service and the need to base the quantitative aspects of the description on information from good quality clinical trials or meta-analyses. There are challenges in converting relatively complex trial-related definitions of outcomes (e.g. hypoglycaemic episodes) into descriptions easily understood by a range of patients.
No attempt was made to determine whether there was a disparity between reported and actual willingness to pay. This issue remains one of the major challenges for research in this area.
As a result of this study, a committee in charge of the Australian guidelines, was set to reconsider whether the policy of discouraging the use of cost-benefit analyses will continue in future.
Source of funding Supported by a grant from Eli Lilly Australia Pty Limited, Sydney, Australia.
Bibliographic details Davey P, Grainger D, MacMillan J, Rajan N, Aristides M, Dobson M. Economic evaluation of insulin lispro versus neutral (regular) insulin therapy using a willingness to pay approach. PharmacoEconomics 1998; 13(3): 347-358 Other publications of related interest Davey P, Grainger D L, MacMillan J, et al. Clinical outcomes with insulin lispro compared with regular insulin: a meta analysis. Clin Ther 1997;19(4):656-674.
Indexing Status Subject indexing assigned by NLM MeSH Australia; Diabetes Mellitus /drug therapy /economics; Humans; Hypoglycemic Agents /administration & Insulin /administration & Insulin Lispro; derivatives /economics; dosage /analogs & dosage /economics AccessionNumber 21998008148 Date bibliographic record published 31/10/2001 Date abstract record published 31/10/2001 |
|
|
|