|
nif:isString
|
-
We conducted a cost-effectiveness analysis with Markov modelling from payers’ perspective. We conducted a literature survey to define the alternative immunisation programmes and to construct the model. Studies pertaining to epidemiology and prognosis of relevant diseases caused by S. pneumoniae in Japan’s setting were accessed from PubMed database, Igaku Chuo Zasshi (Japana Centra Revuo Medicina) database, Ministry of Health, Labour and Welfare (MHLW) Grant System, and annual statistic reports published by the government. Igaku Chuo Zasshi, a Japanese medical bibliographic database, which contains over 10 million citations originating from Japan, comprehensively covers articles published in Japanese-language medical journals. Due to insufficient evidences from Japan, overseas’ reports from PubMed, Medline, The Cochrane Database of Systematic Reviews, Health Technology Assessment database, and The NHS Economic Evaluation Database regarding vaccine effectiveness, utility weights to estimate quality adjusted life year (QALY) and economic evaluation related to vaccines were used instead. PPSV–23 programmes and inclusion of PCV–13: The target population of the immunisation programmes to be evaluated are those aged 65 and older in 2014. In evaluating the efficiency of different PPSV–23 immunisation programmes, we set three different strategies with different ages to receive a shot of subsidised vaccine, namely: (1) current PPSV–23 strategy, (2) 65 to 80 (as “65–80 PPSV–23 strategy”), and (3) 65 and older (as “≥65 PPSV–23 strategy”). Age-specific populations were from demographic data [10]. Current PPSV–23 strategy served as a comparator of the other two strategies. In 65–80 PPSV–23 strategy, those who aged over 80 were not eligible to the immunisation programme, which means these individuals will only follow the transition probabilities assigned to the corresponding ages without any vaccine effectiveness on reducing any S. pneumoniae-related diseases. Vaccine uptake rates were assumed at 50.4% for all strategies, which was the same with the coverage rate of seasonal influenza vaccine in 2013 [11]. In order to investigate the cost-effectiveness of PCV–13 inclusion in the list for single-dose pneumococcal vaccine national immunisation programme, we made variations on the share of PCV–13 between the two pneumococcal vaccines from 10% to 100% with 10% interval, because it is unknown how doctors, vaccinees, and municipalities will choose between PPSV–23 and PCV–13. Ten levels of diffusion of PCV–13 were compared with current PPSV–23 strategy. Only single-dose subsidy was analysed and not the sequence of PCV–13/PPSV–23 or PPSV–23/PCV–13, this is mainly due to PPSV–23 immunisation being a newly-launched programme in Japan [12, 13]. We reserve the evaluation of the cost-effectiveness of uptaking the two vaccines in the future research so as to delineate and emphasize on the main purpose of the study.
A Markov model of courses followed by the cohort under consideration was constructed based on epidemiological data, vaccine effectiveness and models from previous studies. Seven mutually-exclusive health states were modelled: health (without any S. pneumoniae-related diseases), bacteremia without pneumococcal pneumonia, bacteremia with pneumococcal pneumonia, meningitis, CAP caused by S. pneumoniae, neurological sequelae and death of or other than the related diseases (Fig 1). A Markov cycle for each stage was set at one year with a cohort timeframe of 15 years after being vaccinated. We assumed all the individuals who survived until the timeframe age have a life expectancy of the Japanese population [14]. Adverse effects associated with vaccination of PPSV–23 and PCV–13 were not considered, since they were mild or moderate in severity [15–17]. Considering that all transition states did not occur simultaneously at the end of each cycle, while in reality, most kinds of transitions typically occur gradually throughout a time interval (on average, half-way through), we implemented a half-cycle correction in estimating the incremental cost-effectiveness ratios (ICERs) of the programmes [18]. The half-cycle correction is implemented by using one-half of every state’s incremental reward in model’s initial and final reward.
Figure data removed from full text. Figure identifier and caption: 10.1371/journal.pone.0139140.g001 Markov Model. Outcomes in terms of QALY were estimated by assigning transition probabilities and utility weights from literature. We estimated the 5-year age-specific incidence rates using the following: (1) annual IPD incidence rates among persons age 65 and over (2.41 per 100,000) [19], (2) IPD distribution by age [19], and (3) demographic data [10]. NPP annual incidence rates were estimated as incidence of CAP times proportion of S. pneumonia-caused CAP at 17.2% [20]. CAP incidence rates, 10.7 and 42.9 per 1000 person-years for persons who were aged 65–74 and aged ≥75, respectively, were from a 3-year prospective hospital-based surveillance [21]. Proportions of bacteremia with/without pneumococcal pneumonia, and meningitis among IPD cases were from the Infectious Agents Surveillance Report (IASR) [19]. Ubukata’s results of IPD case-fatality rates and proportion that results in neurological sequelae among IPD cases and NPP cases were used in the study [22]. NPP case-fatality rate was from Ishida et al.’s study, which reported the rate among patients with positive urinary antigen test of S. pneumoniae-related pneumonia [23]. Deaths of causes other than the above diseases were taken from the vital statistics [24]. Utility weights used to calculate QALY were assumed based on a study by Smith et al. [25]. Average lengths of hospital stay were from published government data [26]. All these data are shown in Table 1.
Table data removed from full text. Table identifier and caption: 10.1371/journal.pone.0139140.t001 Model inputs. aOn Markov model, transition probabilities from health state A to health state B by ages were calculated as followsFrom “Health” to “Bacteremia without pneumococcal pneumonia” = Annual incidence rate of IPD × Bacteremia without pneumococcal pneumonia among IPD casesFrom “Health” to “Bacteremia with pneumococcal pneumonia” = Annual incidence rate of IPD × Bacteremia with pneumococcal pneumonia among IPD casesFrom “Health” to “Meningitis” = Annual incidence rate of IPD × Meningitis among IPD casesFrom “Health” to “Non-bacteremic pneumococcal pneumonia” = Annual incidence rate of CAP × CAP caused by S. pneumoniae Vaccine effectiveness of PPSV–23 in reducing IPD incidence rate was cited from a Cochrane Review report [30]. Results from meta-analysis show that the use of PPSV–23 to prevent vaccine serotype IPD in adults of high-income countries, was at 82% (69%-90%), while its effectiveness in reducing non-IPD was inconsistent. We assumed that the effectiveness in reducing non-IPD to be 0% [30]. PCV–13 effectiveness in reducing vaccine-serotype IPD and non-invasive vaccine-type CAP, 75.0% and 45.0%, respectively, were from a randomised placebo-controlled trial study [31]. We assumed the effectiveness of both vaccines reduce by age of vaccination and by years after being vaccinated. Extent of reduction was proportional to the effectiveness used in Smith et al.’s study [25]. All these data are shown in Table 2. The vaccine serotypes causing IPD among elderly were 60.0% and 46.0% for PPSV–23 and PCV–13 [19], respectively, those for NPP were 62.7% and 49.3% [13] (Table 1).
Table data removed from full text. Table identifier and caption: 10.1371/journal.pone.0139140.t002 Data used to estimate vaccine effectiveness (VE) and VEs used in the model. In this study, costs borne by government, municipalities, vaccinees, patients and third party payers were considered, while advertising costs borne by manufacturers were left unaccounted. It is obvious that when a new product enters a market, which was monopolised by the other product, the manufacturers of both products will invest a lot to compete for the share in the market. Since the decision maker, MHLW Vaccine Committee, was only interested in costs borne by the aforementioned payers, we omitted the inclusion of this cost. Non-direct medical costs related to the immunisation programme were not included, because the programme was built within the public health services routine [32]. The amount of direct payments to health care providers by these entities was estimated as costs. Cost items were identified along the decision tree and Markov model. We used the literature along with some assumptions to estimate necessary data. Age-specific treatment costs of per case of bacteremia with/without pneumococcal pneumonia and pneumonia were estimated from Status Survey on Medical Care Benefits [26]. Cost per case of meningitis was assumed to be twice the cost per case of bacteremia, while cost of sequelae was assumed ¥1,500,000 (US$13,636) per case per year [27]. One vaccine shot was assumed at ¥8,116 (US$ 74: US$1 = ¥110) for PPSV–23 and ¥10,776 (US$98) for PCV–13, which were the sum of vaccine price (¥4,737 or US$43 for PPSV–23, ¥7,200 or US$65 for PCV–13) [28, 29], doctor fee and technical fee for PPSV–23 and PCV–13, respectively. All cost data are shown in Table 1.
Outcomes and costs were discounted at a rate of 3% [32].
One-way sensitivity analyses and probabilistic analyses: We performed one-way sensitivity analyses on two pairs of comparisons. The first pair is ≥65 PPSV–23 strategy vs. current PPSV–23 strategy, which is to appraise the stability of ICERs with the assumptions made in our economic model, and to explore the impact of each variable relative to each other when the subsidy of the immunisation is limited to PPSV–23. The second pair is PCV–13 strategy vs. current PPSV–23 strategy, which is to appraise the same issues when PCV–13 is also subsidised by current immunisation programme. The lower limits and upper limits used in sensitivity analyses were ±50% for costs variables and ±20% for probabilities and utilities. We also conducted two sets of 1000 Monte Carlo simulations on ≥65 PPSV–23 strategy and 65–80 PPSV–23 strategy vs. current PPSV–23 strategy, i.e., probabilistic analyses, for which all data were assumed to have an equilateral triangle distribution corresponding to the range tested in one way sensitivity analyses. Triangular distribution was used because of the insufficiency of information about distributions. This distribution has been theoretically proven as both simple and efficient, which can serve as a proxy for beta or other distributions in risk analysis [33–35].
|
![[This page as RDF]](http://data.gesis.org/somesci/static/rdf-icon.gif){kind=icon}