Application of the screening and indirect cohort methods to evaluate the effectiveness of pneumococcal vaccination program in adults 75 years and older in Taiwan

Background The Taiwanese national 23-valent pneumococcal polysaccharide vaccine (PPV23) program in adults ≥75 years of age and the 13-valent pneumococcal conjugate vaccine (PCV13) program for children were implemented in 2008 and 2013, respectively. In this study we evaluated PPV23 vaccine effectiveness (PPV23VE) in the elderly, with regard to both direct protection from the vaccine itself and the indirect protection conferred by PCV13 immunization in children. Methods The incidence of invasive pneumococcal disease (IPD) in Taiwan from July 2008 to June 2016 was collected from IPD surveillance data. A comparison of IPD incidence with a nationwide vaccination registry allowed an estimation of PPV23VE by the screening and indirect cohort methods. Results The incidence of IPD in adults ≥75 years of age ranged from 13.9 per 100,000 inhabitants during the period July 2008–June 2013 to 10.4 per 100,000 inhabitants between July 2013 and June 2016 (relative risk [RR]: 0.75; 95% confidence interval [95% CI]: 0.67–0.85). According to the screening method, PPV23VE against death within 30 days of IPD onset, all IPD, and PPV23-serotype IPD was 32.5% (95% CI: 17.5–44.7%), 33.9% (95% CI: 25.2–41.5%) and 43.4% (95% CI: 34.4–51.2%), respectively. PPV23VE with the indirect cohort method was 39.0% (95% CI: 15.5–55.9%) for all PPV23 serotypes and 71.5% (95% CI: 44.2–85.4%) for 11 serotypes included in PPV23 but not in PCV13. During the period July 2008–June 2012, PPV23VE against PPV23-serotype IPD was 55.1% (95% CI: 27.2–72.3%). Conclusions PPV23 is able to prevent IPD and 30-day fatality in adults 75 years of age and older due to a combination of direct effects from PPV23 and indirect effects from PCV13. It might confer higher protection against PPV23-serotype IPD before the introduction of PCV13 program in children. Supplementary Information The online version contains supplementary material available at 10.1186/s12879-020-05721-0.


Background
Invasive pneumococcal disease (IPD), defined as the isolation of S. pneumoniae from a normally sterile body site, such as the blood, cerebrospinal fluid or pleural effusion, poses significant health threats to young children (< 5 years of age), older adults, and individuals with chronic medical conditions [1,2]. In the USA, a national pneumococcal conjugate vaccine (PCV) program for children was introduced in 2000, which was earlier than most countries worldwide [3]. The pediatric population was chosen because the incidence of IPD was higher in children under the age of 5 years (71.8 per 100,000 inhabitants) than in adults ≥ 65 years (57.6 per 100,000 inhabitants) [4]. However, by 2016, the main disease burden of IPD in the USA had shifted from children < 5 years (8 per 100,000 inhabitants) to adults ≥ 65 years (24 per 100,000 inhabitants) [5]. Similar findings were determined in Taiwan from the pre-to the post-PCV immunization era [3,6]. Age-related changes in IPD epidemiology over time indicate the need to evaluate the impact of pneumococcal immunization programs and vaccine effectiveness (VE) both in children and in older adults [3,7,8].
National 23-valent pneumococcal polysaccharide vaccine (PPV23) and 13-valent PCV (PCV13) vaccination programs were started in Taiwan in 2008 and 2013, respectively [3,6,9]. Thus, any evaluation of the protective effects of PPV23 in the elderly must take into account both direct protection from PPV23 vaccination and indirect protection from PCV13 immunization in children [10].
While evidence-based recommendations form the cornerstone of the vaccination policy of the World Health Organization (WHO) and of many developed countries, such data are scarce in Africa and Asia [7,11,12]. Available systematic reviews and meta-analyses have concluded that PPV23 confers protection against IPD but the duration of protection is unclear [11,[13][14][15][16]. We therefore investigated the PPV23 program in older adults in Taiwan over a study period of 8 years, from the pre-to the post-PCV13 immunization era.

Methods
The PPV23 vaccination program and vaccine coverage A national PPV23 vaccination program aimed at the elderly (≥75 years), with vaccines donated by the Formosa Plastic Group, was implemented in 2008 [9]. In addition, during the study period seven counties/cities (Tainan, Yunlin, Taichung, and Lianjiang counties, and the cities of Taichung, Tainan, and Chiayi) introduced local, publicly funded PPV23 vaccination programs that provided free PPV23 immunization to residents 65-74 years of age.

Data sources
Data on IPD cases in patients of all ages in whom disease onset was between July 1, 2008, and June 30, 2016, were obtained from the national IPD surveillance system, a hospital laboratory-and case-based passive surveillance system for monitoring IPD established on October 15, 2007, by the Taiwan Centers for Disease Control [2,17]. The IPD surveillance database contains demographic and clinical data, including IPD onset date and high-risk medical conditions (HRMCs, such as immunodeficiency/cancer, chronic obstructive pulmonary disease, congenital heart disease, splenectomy/asplenism, neurological disease, organ transplantation, congenital metabolic disorders, and other major illnesses) [2].
The PPV23 vaccination date was obtained from the National Immunization Information System (NIIS) database, described in previous studies [2,18,19]. The national vaccine registry is primarily designed to collate childhood vaccination data into a single web-based repository. However, it has been extended to include the registration of publicly funded PPV23 vaccination in the elderly and voluntary, self-paid childhood and adult vaccination. By linking the NIIS to the National Household Registration System, which includes all citizens with identifiers, we were able to calculate the coverage rate of PPV23 immunization in Taiwan for vaccine-targeted age groups.

Statistical analysis
Incidence rates of IPD were calculated per 100,000 inhabitants, and specific incidence rates by age groups (≤5, 6-64, 65-74, and ≥75 years) and vaccine serotype. Incidence rate of serotype 19A-IPD was specifically characterized because it emerged after 7-valent PCV introduction and tends to result in more complicated pneumonia with empyema [6]. The rates were compared using relative risk (RR) and the 95% confidence interval (95% CI). Age-specific data on inhabitants in Taiwan were obtained from the Taiwan National Household Registration on a yearly basis. The Cochran-Armitage test was used to assess the trends in annual IPD incidence. Individuals were considered to be vaccinated if their PPV23 vaccination date was ≥14 days before IPD onset. We excluded those received neither 2 doses of PPV23 nor PCV13 plus PPV23 from the study. Differences between vaccinated and unvaccinated patients were estimated using a chi-squared test or Fisher's exact test to compare proportions and by a linear regression model to compare continuous variables.
PPV23VE was calculated using two methods: the screening method and the indirect cohort (Broome) method. The screening method, described by Farrington [10,20,21], is based on the comparison of the proportion of vaccinated cases with the proportion of the vaccinated population [21], and by its assumption in nature, instead of data on a control group, data on the whole population are used for contrast with vaccine coverage in the cases [22]. The VE was expressed as VE where Pc=the proportion of cases who have been vaccinated and Pp =the proportion of the target population who have been vaccinated. Using logistic regression models, VE obtained by the screening method could control for confounding variables of age group (75-84 and all 85+) and sex when data on the vaccination coverage in each subgroup was available.
On the other hand, the indirect (Broome) cohort design used IPD cases caused by PPV23 vaccine types (VT) as the case group and IPD cases caused by non-PPV23 serotypes (ST) as the control group (non-cases, the comparison group) [10,23]. The basic assumption in the indirect cohort design is that PPV23 vaccine provides no protection against and does not increase the risk of IPD caused by non-PPV23 ST. [10,[23][24][25]. VE was estimated by comparing the vaccination odds (compared to no PPV23 vaccination) of cases with controls and calculated as (1− odds ratio) × 100% [10,23]. Potential confounders, including sex, age, HRMC, and year of symptom onset, were adjusted by logistic regression. The statistical power of indirect cohort method would decrease as the vaccine coverage increases (> 50%) and fewer VT cases occur [23]. VE was also estimated for IPD caused by 1) 11 serotypes included in PPV23, but not found in PCV13 (PPV23-non PCV13 VT); 2) each serotype included in PPV23 that had been identified in at least 30 cases, in which other PPV23 vaccine serotypes were excluded for analysis. VE in preventing PPV23-serotype IPD, calculated according to the indirect cohort method, was also expressed as the number needed to vaccinate per case prevented [10,24]. To estimate VE according to different intervals after PPV23 vaccination, the interval between IPD onset and the date of vaccination was categorized as ≤ 1 year, > 1 and ≤2 years, > 2 and ≤ 3 years, > 3 and ≤ 4 years, > 4 and ≤5 years, and ≥ 5 years. All analyses were conducted using SAS software (ver. 9.4; SAS Institute, Cary, NC, USA).

Ethical statement
The Taiwan CDC approved the protocol of this study and waived the requirement for written informed consent because of the study's retrospective design and the use of data from administrative databases, thus, involving minimal risk to study participants.

Characteristics of IPD in adults ≥ 75 years
The incidence of IPD in adults ≥75 years of age varied from 16.11 per 100,000 inhabitants during July 2008-June 2009 to 9.17 per 100,000 inhabitants during July 2015-June 2016 (p for trend <.0001).

Discussion
For countries with long-term laboratory-based systems for monitoring IPD, linking IPD surveillance with valid pneumococcal immunization records facilitates the evaluation of vaccination programs pre-and postimplantation from a public health perspective [1]. In determinations of VE, the results of effectiveness studies might be at risk of bias due to the impact of patient health and the short study period, as was the case in a previously published nationwide study of PPV23 effectiveness in Taiwan. In contrast, our study used data from subsequent years in similar adult populations and drew on different observational methodologies [9,16]. The results showed that with respect to IPD and 30-day fatality, PPV23VE was lower than expected when the study period was extended to 8 years. Based on observational studies, the estimated PPV23VE against IPD in older adults or adults with conditions associated with an increased risk of IPD was 27-76% [9,11,13,16,24,26,27]. Previously reported estimates of VE have differed, most likely because of the different methods used and the different study periods in the estimations [14,27]. The point estimate of PPV23VE in this study was within the 95% CI of the estimated VE reported in a Cochrane review of non-random controlled trials in adults (VE = 52, 95% CI: 39-63%) [13].
The decline in IPD among adults ≥ 75 years from preto post-PCV13 immunization era is likely due to a combination of direct effects from PPV23 and indirect effects from PCV13, which are epidemiologically challenging to tease apart. Therefore, we applied the indirect cohort method to evaluate PPV23VE during the period July 2008-June 2012 when national PCV13 program was not introduced in children. PPV23VE for elderly seemed higher in the period of July 2008-June 2012 (VE=55.1, 95% CI: 27.2-72.3%) than in the period of July 2008-  June 2016 (VE=39.0, 95% CI: 15.5-50.9%). We speculated that as the indirect protection emerged following PCV13 introduction, PPV23 unvaccinated elderly could benefit from lowering their risk of acquiring IPD caused by 12 serotypes common to PCV13 and PPV23, which might bias the PPV23VE estimation to be lower. On the other hand, VE waning since time of vaccination should possibly be considered.
In addition, in our study, PPV23VE was higher against IPD caused by 11 serotypes that included in PPV23 but not in PCV13 (PPV23-non PCV13 VT) than against PPV23 serotype-IPD, as also found in a Spanish study [10]. Such findings were observed using either the screening method or the indirect cohort method. This difference may be due to a differential effectiveness against different serotypes [10,28]. However, other possible causes remain to be explored in future studies [10,23,28,29].
The validity of VE estimation using the screening method could be influenced by the completeness of PPV23 immunization recorded in the NIIS. Although the accurate PPV23 vaccination coverage for IPD cases and elderly population might be greater than 32.8 and 41.9%, respectively, it may not result in very low or very high Pp and Pc to bias VE estimation [30,31]. PPV23VE estimates by the indirect cohort method would possibly bias if the PPV23 vaccinated and unvaccinated individuals are not at the same risk of non-PPV23 ST infections [10,23]. Unlike PCV-induced serotype replacement, PPV23 would not drive the increase of non-PPV23 ST among vaccinated and unvaccinated elderly.
The adjusted VE determined by the screening method was slightly higher than the estimated VE obtained using the Broome method, perhaps because the screening method does not allow for the control of confounding factors, which can result in an overestimation of VE [21]. Moreover, the study population applied in the Broome method consisted of all patients who had developed IPD (case-case comparison approach) [32], in whom the proportion of underlying disease may have been higher than in community-dwelling elderly. Differences in the immune responses to vaccination of the case-case population compared to the communitydwelling elderly population may lead to a lower VE estimate [16,33].
Following the 2006 recommendation by the WHO of a routine PCV-based immunization program in children, the IPD incidence caused by PCV VT gradually declined not only in vaccinated (direct protection) but also in unvaccinated (indirect protection) age groups [3,34,35]. From before (July 2008-June 2013) to after (July 2013-June 2016) the implementation of a children's PCV13 program in Taiwan, the highest age-specific annual IPD incidence shifted from age 2-4 years to ≥ 75 years, with a decreasing trend in the incidence of IPD related to PCV13 VT across all age groups. Previous studies reported an increase in IPD, due to non-PCV serotypes or serotype replacement, roughly 3-4 years after the introduction of a PCV program [36][37][38]. The increasing trend in the incidence of non-PCV13 ST IPD determined in this study involved all age groups except adults ≥75 years, i.e., the target population of the national PPV23 program. Whether serotype replacement occurs over different time intervals in different age groups or is influenced by the different serotype distributions among age groups, especially for those covered by the PPV23 vaccination, remains to be investigated. Long-term surveillance of IPD, nasal carriage, and non-bacteremic pneumococcal pneumonia will be crucial in the monitoring of serotype replacement and in ascertaining whether pneumococcal vaccines offer direct or indirect protection against the incidence of disease over the long term [36]. Vaccine protection at the population level can be rapidly estimated using the screening method whereas the duration of PPV23 protection at an individual level could be estimated by a stratified analysis of time since vaccination using the indirect cohort method [10,11,21]. In this study, the change in PPV23VE since the time of vaccination, measured at an interval of every 1-year between IPD onset and the date of vaccination, could be considered a diagnostic indication of waning VE [10,15,39]. The wide confidence interval and biased point estimate of VE in the subgroup of the interval of > 4 and ≤ 5 years could be possibly explained by the small sample size [40]. A study from England and Wales reported a PPV23VE of 48% against IPD within 2 years of vaccination for adults ≥65 years, as determined by the indirect cohort method, but VE waned and became insignificant beyond 5 years [28]. In the effectiveness of pneumococcal vaccination against community-acquired pneumonia, acute myocardial infarction and stroke (CAPAMIS) study, effectiveness estimates became higher and significant after patients vaccinated > 60 months previously were excluded [41]. The differential VE against PPV23serotype IPD observed over time in our study was in agreement with a previous report but the lack of significance of the 2-to 5-year interval may have been due to the lower statistical power resulting from stratification of the variable "time since vaccination" [10].
PPV23 is a T-cell-independent vaccine that lacks a mechanism for long-term boosting of the immune response [42]. While PPV23 revaccination may result in a significant and sustained antibody responses in adults, including the elderly [43,44], it is recommended only for those with an increased risk of IPD and no sooner than 5 years after the first dose [45]. However, insufficient data regarding clinical benefit, the degree and duration of protection, and safety have hindered a routine recommendation of revaccination [46]. In the USA, as in Taiwan, a single dose of PPV23 is recommended for all adults ≥65 years of age regardless of the previous history of PPV23 [3,46]. In the elderly and in risk groups in middle-high income countries, PCV13 and PPV23 immunization is recommended to confer better and longer protection [42,47]. In fact, in many countries, PPV23 is recommended as an effective and/or cost-effective vaccine covering a broad array of the serotypes implicated in IPD in older adults [10,45,48].
Our study had several limitations beyond its observational design and the residual confounders. First, although the coverage of hospitals enrolled in IPD surveillance in Taiwan should have been 100%, detection bias may have occurred for cases identified using this passive surveillance system, thereby affecting the representativeness of the data [2,17]. Second, PPV23 vaccine records in the NIIS database were prospectively collected after the endorsement of a publicly funded PPV vaccination program for elderly individuals. Information on missing vaccination status was not determined by the occurrence of IPD outcome. Therefore, the probability of misclassification of vaccinated to be unvaccinated might be non-differential between cases and controls in the indirect cohort method. Third, IPD surveillance has been started with the implementation of the national PPV23 vaccination program in the elderly. We could not evaluate the impact of PPV23 program from pre-to post-PPV23 vaccination program by changes of IPD incidence. Fourth, influenza vaccination status is an important covariate that should be considered in evaluations of PPV23VE in the elderly [49]. However, the NIIS database contains limited annual individual influenza vaccine records and it was not possible to determine seasonal influenza status as covariate information for the elderly population in our study.

Conclusions
In conclusion, our results demonstrated a decrease in IPD incidence among all age groups across eight seasons covering the pre-to post-national PCV13 immunization era in Taiwan, accompanied by an increased trend of non-PCV13 ST IPD. The decline in IPD among adults ≥ 75 years is likely due to a combination of direct effects from PPV23 and indirect effects from PCV13. Using the indirect cohort method, PPV23 did confer moderate protection against PPV23-serotype IPD in adults ≥75 years before national PCV13 immunization program implemented in children.