- Email: [email protected]
A model to estimate the probability of hepatitis B- and Haemophilus influenzae type b-vaccine uptake into national vaccination programs
Vaccine 18 (2000) 2223±2230
www.elsevier.com/locate/vaccine
A model to estimate the probability of hepatitis B- and Haemophilus in¯uenzae type b-vaccine uptake into national vaccination programs Mark A. Miller a,*, W. Dana Flanders b a
Children's Vaccine Initiative Secretariat, c/o World Health Organisation, 20 Avenue Appia, 1211, Geneva, Switzerland b Emory University, Atlanta, GA, USA Received 16 March 1999; received in revised form 17 November 1999; accepted 7 December 1999
Abstract Most countries have been slow to adopt new vaccines into national vaccination schedules, despite recommendations from global multi-lateral agencies. Characteristics of countries that have adopted hepatitis B (HB) vaccine were analysed and used to formulate a logistic regression model. The model was applied to country-speci®c data to predict HB and Haemophilus in¯uenzae type b (Hib) vaccine uptake. The greatest predictors of HB uptake were coverage rates of other vaccines, vaccine cost relative to the economy, and perceived disease burden. The logistic regression model's probability estimate of vaccine uptake agreed well with observed data for HB and Hib, (c-statistic 85 and 82%, respectively). Application of this model to other antigens may aid in predicting potential national markets to better plan new vaccine supply and demand. 7 2000 Published by Elsevier Science Ltd. All rights reserved. Keywords: Hepatitis B; Haemophilus in¯uenzae type b; Vaccines; Economics
1. Introduction After the success of smallpox eradication, the World Health Organisation's (WHO) Expanded Programme on Immunisation (EPI) has helped to introduce six basic vaccines, diphtheria±tetanus±pertussis (DTP), Bacille Calmette±GueÂrin, measles, and poliomyelitis, into national vaccination programs throughout the world. In 1988, the EPI recommended yellow fever vaccine to be added to national immunisation schedules for use in countries with endemic disease [1]. The World Health Assembly recommended hepatitis B (HB) vaccine for universal childhood vaccination in 1992 [2]. Most recently, the WHO recommended Hae* Corresponding author. After February 1: Fogarty International Center, National Institutes of Health, Bethesda, MD 20892, USA. Tel.: +1-301-496-1415; fax: +1-301-496-2173. E-mail address: [email protected] (M.A. Miller).
mophilus in¯uenzae type b (Hib) conjugate vaccines for use in routine immunisation programs in countries with a demonstrated disease burden, ``as appropriate to national capacities and priorities'' [3]. Despite these recommendations, many countries have been slow to adopt these vaccines in routine immunisation schedules. Yellow fever vaccine has been available since 1937, yet fewer than 13 of the 45 countries where yellow fever is endemic report routine use of this vaccine (EPI, unpublished data). HB vaccine was ®rst licensed in 1981, however, less than one-half of the world's children born each year receives this vaccine in routine schedules [4]. Despite the availability of Hib conjugate vaccines for almost 10 years, their use is limited to less than 20% of the global birth cohort (J. Wenger, WHO, unpublished data). Numerous factors have been cited for the slow uptake of new vaccines. These include vaccine cost,
0264-410X/00/$ - see front matter 7 2000 Published by Elsevier Science Ltd. All rights reserved. PII: S 0 2 6 4 - 4 1 0 X ( 9 9 ) 0 0 5 6 3 - 0
2224
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
poor infrastructure to deliver vaccines and lack of adequate data on disease burden. However, there has been no systematic quanti®cation of the importance of these factors. With the development of new eective vaccines against enteric disease (rotavirus, cholera, typhoid), pneumonia (pneumococcal conjugate) and meningitis (meningococcal conjugate), estimating potential vaccine demand and quantifying the magnitude of the barriers for their introduction is becoming increasingly important. Using HB vaccine uptake as a model, we quantify the factors that are associated with vaccine uptake into national immunisation schedules and apply a model to estimate the probability of HB and Hib vaccine adoption. 2. Methods The goal of these analyses is to identify and assess factors associated with the adoption of HB vaccines into routine national vaccination programs. Characteristics of countries that have adopted HB vaccine were compared to those countries that have not. Logistic regression models were developed based on country speci®c epidemiologic and economic factors. These factors were then applied to data speci®c to Hib disease and vaccine to assess the model's predictive ability of Hib vaccine uptake.
Table 1 List of variables studied Variable Demographics 1996 Surviving infants and total population Geopolitical region WHO Region Economic 1996 Per-capita GDP World Bank income group Program factors Vaccine cost (per dose) DTP vaccine coverage Health characteristics Life expectancy at birth HB surface antigen prevalence Deaths in current birth cohort due to HB Years of life lost per birth Unit case treatment cost Integrated factors Vaccine program cost Deaths prevented per 1000 infants Life-years saved per 1000 infants Treatment costs prevented per 1000 infants a
UNPD±United Nations Population Division. WB±World Bank. c WHO±World Health Organisation. d CVI±Children's Vaccine Inititative. b
Data source UNPDa WBb WHOc WB WB CVId WHO WHO WHO CVI CVI CVI CVI CVI CVI CVI
The HB and Hib models were based upon countryspeci®c epidemiologic and economic data previously described [4]. Of the 215 countries or territories that report data to the WHO, 36 were excluded due to data limitations. Of those excluded, all have populations less than 150,000 and most are island nations. Countries were characterised as having no routine scheduled HB vaccination program (either infant/childhood or adolescent), planning a HB vaccination program or current HB vaccination program as of October, 1998. 2.1. Data collection Published data pertaining to HB vaccination programs, which include country speci®c demographics, economic data and data speci®c to their national vaccination program, served as variables for the model. Speci®c variables and data sources are speci®ed in Table 1. Factors considered as potential predictors include descriptive data of each country's demographic features (total population, birth cohort) [5] and economy as de®ned by per-capita gross domestic product (GDP) [6] and geo-political category as de®ned by the World Bank [7]. Program factors assessed for each country included the unit cost of vaccination, including vaccine and administration; and the coverage rate for DTP vaccine. Measures of health characteristics of each country assessed included: the prevalence of HB surface antigen (hbsag), an indicator of hepatitis B endemicity, and the estimated annual incidence of deaths and years of life lost per infant (YLL) due to the speci®c vaccine preventable disease. The methodology for the calculation of mortality and YLL are summarized in a previous study [4]. Integration of these variables was also assessed. For example, various health outcomes, such as the reduction in incidence of death and years of life lost through HB vaccination were determined by integrating disease burden and coverage rates. In addition, treatment costs prevented is based on integrating disease burden, coverage and economic factors. Country speci®c determinations were adjusted for the size of the birth cohort. The methodology for the derivation of these characteristics has been previously described [4]. 2.2. Analyses In exploratory analyses, the mean and standard deviation for each potential predictor variable were compared between countries that adopted HB vaccine into its national vaccination schedule with the corresponding measures among countries that had not adopted such a policy. We use the Wilcoxon rank sum test [8]
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
to test for dierences between the distribution of continuous variables in countries adopting a vaccination policy versus other countries. Logistic regression was used to study further predictors of each country's decision to adopt a national vaccination program for HB. Each variable was grouped into quartiles (or other natural groupings) to assess the general pattern of the relationship between vaccination and the variable. If the exploratory analysis was consistent with a monotonic dose±response, the variable was used as a continuous variable in subsequent logistic regression analyses. We identi®ed a predictive model for adoption of vaccination by ®rst using a priori considerations and the exploratory analyses to select key variables. These variables were further assessed in multivariate analyses for their predictive ability while, simultaneously controlling for the other variables. To illustrate the association between each variable and vaccination status, we calculated odds ratios (OR) as follows: the OR represents the odds that a country at the mean of the highest quartile would adopt vaccine relative to the corresponding odds for a country in the lowest quartile (OR=exp[b(x4ÿx1)]; where b=estimated regression coecient, x4=the mean value of the independent variable for countries in the highest quartile of that variable and x1=the corresponding mean for the lowest quartile).
2225
2.3. Regression diagnostics We screened for potential collinearity by calculating correlation coecients. The logistic regression model was then applied to the country-speci®c data for HB and Hib to determine each country's probability of vaccine uptake. These estimated probabilities were compared to the observed proportion of countries that had committed to adopt HB into their routine infant immunization schedule as of October 1998. We calculated the area under the receiver operating characteristic curve (c-statistic) to indicate the discriminatory ability of the model to distinguish countries that vaccinated from those that did not [9]. We used the Hosmer±Lemeshow goodness of ®t test to assess model calibration [9]. 3. Results Of the 179 countries studied, 75 (42%) have adopted HB into their routine vaccination schedule as of October 1998, (Fig. 1). The percentage of countries using HB vaccine varied regionally from 11% of countries in the World Health Organisation's (WHO) African region to 79% in the Western Paci®c region (Table 2). The percentage using HB also varied by income group
Fig. 1. Model prediction of the probability of HB vaccine uptake and countries with current routine use. The estimate of the probability of adoption, determined by the logistic regression, is proportional to the degree of shading for each country. The circles indicate countries that are currently using HB vaccine.
2226
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
Table 2 Frequency of adoption of vaccination policy, by geopolitical and economic characteristics Variable WHO Region African American Eastern Mediterranean European Southeast Asian Western Paci®c World Bank Income Group Classi®cationa Low (GNP R $785) Lower middle ($785 < GNP R $3125) Upper middle ($3125 < GNP R $9655) High (GNP > $9655) World Bank Classi®cation China Established market economies Former socialist economies of Europe India Latin America and Carribean Middle-east Crescent Sub-Sahara Africa Other Asia and Paci®c Islands a
Total number of countries
Number vaccinating (percentage of total in parentheses)
44 36 23 47 10 19
5 (11.4%) 9 (25.0%) 16 (69.6%) 25 (53.2%) 5 (50.0%) 15 (79.0%)
48 49 44 38
4 (8.3%) 24 (49.0%) 19 (43.2%) 28 (73.7%)
1 25 23 1 34 26 43 26
1 (100.0%) 17 (68.0%) 9 (39.1%) 0 (0.0%) 7 (20.6%) 18 (69.2%) 5 (11.6%) 18 (69.2%)
Income groups are classi®ed based on per capita gross national product (GNP) as measured in US dollars.
and geopolitical characterisation, as de®ned by the World Bank. High-income countries were more likely to adopt HB into routine schedules then low-income countries (73.7 vs 8.3%, OR 8.8, p < 0.0001). Several continuous factors were also associated with HB vaccine uptake (Table 3). Per-capita GDP ($8475
vs 3236) and vaccine cost ($3.96 vs 1.65 per dose) were signi®cantly higher in those countries that routinely vaccinate against HB than those that do not. However, when the vaccine cost was adjusted as a percentage of the country's GDP, the reverse is true (0.17 vs 0.40). Countries that adopted HB had a mean DTP coverage
Table 3 Means of selected factors, by countries that have adopted HB vaccinea Variable Demographic Surviving infants (thousands) Total population (millions) Economic Per-capita GDP (USD) Program factors Vaccine cost per dose (USD) Vaccine cost as percentage of per capita GDP DTP Coverage (percentage) Health characteristics HB surface antigen prevalence (percent) Estimated deaths (per 1000 infants) Estimated years of life lost (per 1000 infants) Unit case treatment cost (USD) Integrated factors Vaccine program cost (per birth) (USD) Deaths prevented by vaccination (per 1000 infants) Estimated life-years saved (per 1000 infants) Estimated treatment costs prevented (per 1000 infants) (USD) a
Mean (SD) in countries not vaccinating
Mean (SD) in countries vaccinating
p-value for dierence in means (Wilcoxon)
727 (2458) 27 (96)
745 (2507) 40 (146)
NS NS
3236 (6481)
8475 (9386)
< 0.0001
1.65 (3.00) 0.40 (0.37) 79.5 (19.8)
3.96 (4.48) 0.17 (0.11) 93.8 (5.8)
< 0.0001 < 0.0001 < 0.0001
6.4 (5.1) 13 (10) 135 (131) 1595 (2790)
5.8 (4.9) 12 (10) 270 (218) 3857 (3825)
NS NS < 0.0001 < 0.0001
9.42 (17.96) 9 (7)
23.49 (26.18) 10 (9)
< 0.0001 NS
102 (114) 230 (496)
239 (193) 1381 (2406)
< 0.0001 < 0.0001
Variables in bold are signi®cantly dierent at the p < 0.0001 level.
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
2227
Table 4 Univariate associations of selected factors with vaccination status, based on logistic modela Variable Economic Per Capita GDP Program factors Vaccine cost (per dose) Vaccine cost/per-capita GDPc DTP coverage Health characteristics HB surface antigen prevalence Deaths (per 1000 infants) Years Life Lost (per 1000 infants) Unit treatment cost Integrated factors Deaths prevented Years life saved Treatment cost prevented
Odds ratiob
95% Con®dence intervals
4.4
2.1
9.4
4.1 39.7 55.1
1.9 6.5 10.4
8.6 241.2 292.6
0.7 0.7 6.9 5.0
0.4 0.8 2.9 2.3
1.5 1.5 16.1 11.0
1.6 9.6 127.4
0.8 3.8 15.5
3.4 23.9 1043.6
a
Bold type indicates signi®cant variables. Odds ratios represent highest quartile relative to lowest quartile in each category. c Lowest quartile relative to highest quartile was compared to yield an OR greater than 1 for comparisons to other variables.
b
greater than countries that have not adopted HB vaccine (93.8 vs 79.5%). Carriage prevalence, as measured by HB surface antigen, did not dier substantially between the two groups, nor did the estimated deaths that would occur in the absence of vaccination. However, life expectancy in a country, when integrated with HB mortality to estimate an incidence of years of life lost due to HB, was strongly associated with adoption of HB vaccine. In addition, countries that spent more on HB treatment per capita were more likely to have adopted HB vaccine. When these factors were combined, countries that have adopted HB vaccine, on average, were paying more per capita for their HB vaccine program ($23.49 vs 9.42) but were saving more life years (239 vs 102) and more in HB treatment costs ($1381 vs 230) per 1000 infants.
larly strongly associated with vaccine adoption (OR 127.4, 95% CI 15.5±1043.6). We identi®ed three characteristics of countries that were strongly associated with vaccine adoption, even after controlling for the other factors. Vaccine cost as a percentage of per capita GDP, DTP coverage, and YLL per 1000 infants were each strongly associated with increased likelihood of adoption of a vaccination policy. Together these three factors strongly predict a country's policy, with a c-statistic of 85%, and provide a good ®t to the observed data as determined by the Hosmer±Lemeshow test. Based on the model coecients and country speci®c data, a map of the probability of hepatitis B uptake was constructed (Fig. 1). Correlation with the outcome is illustrated since countries with a higher predicted probability of vaccine usage tended to actually be using vaccine as of October 1998.
3.1. Logistic regression Results of logistic regression also showed that several factors were strongly associated with adoption of a vaccination program (Table 4). Vaccine adoption was strongly associated with coverage rates of DTP vaccine (OR 55.1 95% CI 10.4±292.6), vaccine cost relative to per-capita GDP2 (OR 39.7, 95% CI 6.5± 241.2) and estimated YLL per infant in the current birth cohort (OR 6.9 95% CI 2.9±16.1). Treatment cost saved through vaccination (which integrates the unit cost of treatment and coverage rates) was particu2 Lowest quartile relative to highest quartile to yield a OR greater than 1 for comparison to other variables.
3.2. Hib The logistic regression model coecients derived from the HB data were applied to the country speci®c data in the Hib database (Hib vaccine cost, GDP, estimated YLL due to Hib, and vaccine coverage rates) to estimate probabilities of Hib vaccine uptake (Fig. 2). The map illustrates agreement of the probability estimates with the observed data. The c-statistic was 82%, indicating that the model could distinguish between countries that adopted vaccines from those that did not.
2228
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
Fig. 2. Model prediction of the probability of Hib vaccine uptake and countries with current routine use. The estimate of the probability of adoption, determined by application of the parameter estimates from the HB model to the Hib data, is proportional to the degree of shading for each country. The circles indicate countries that are currently using Hib vaccine.
3.3. Estimated demand of vaccine doses
4. Discussion
Based on the models' estimate of a country's probability of adopting HB or Hib vaccines, we have calculated the number of required vaccine doses in each category (Table 5). For example, assuming equivalent rates of vaccine use to DTP coverage, 21 million doses would be required annually in countries that have a >75% probability for vaccine uptake. An additional 16 million doses would be required for countries with a >50 to 75% probability of Hib uptake.
Our results highlight the relative importance of the national commitment to the vaccination programs (coverage rates of DTP) and economic indicators relative to disease burden estimates associated with vaccine uptake. Surprisingly, carriage prevalence and number of deaths due to HB did not correlate with vaccine adoption, however, when years of life lost were incorporated, a disease burden correlate was found. We speculate that countries where average life expectancy
Table 5 Distribution of the number of HB and Hib doses required per annum based on the model's estimate probability of vaccine adoptiona Total doses of vaccine (millions) required by countries in each category Estimated probability of vaccine uptake
HB
Hib
0±25% > 25±50% > 50±75% > 75% Total
151 19 54 136 361
333 4 16 21 374
a
Dierences in the total doses result from omission of smaller countries in the HB dataset.
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
is less than 50 are less likely to recognise deaths due to HB, which occur at mid- to late-stage of life, to be a major public health problem relative to those countries whose populations have longer life expectancies and, consequently, years of life lost. Interestingly, the combined variable of treatment costs saved through vaccination, which incorporates coverage, treatment expenditures and disease burden, had the strongest association. The model predicted quite well the probability that a country had adopted HB and Hib vaccines, as re¯ected in the ®gures. For HB, the model did not correctly predict adoption for a few countries, almost exclusively in Latin America, Central Asia and Scandinavia. The low use in Scandinavia may be attributed to the relatively low incidence of disease in the native population and to countries which have an alternative HB vaccination policy of targeting high-risk groups (i.e. infant immunisation in immigrants). Although all ®ve Central Asian nations of the former Soviet Union were estimated to have greater than 50% probability of using HB, only two countries had done so. The recent decentralisation of government health services may have negatively impacted on the introduction of new vaccines. Finally, in Latin America, much of the vaccine policy and purchase is co-ordinated by the Pan American Health Organization (PAHO) which may have impacted on the adoption of vaccine into routine schedules. Some countries in this region have introduced HB vaccine to target high-risk populations in certain geographical regions (e.g. populations in the Amazon). In 1999, due to centralised purchasing, numerous countries in the PAHO region will adopt HB and Hib vaccine, further supporting the model's probability estimates. 4.1. Utility of model Ð market forecasts and demand estimates By categorising the countries by their probability of vaccine uptake and linking it with data of national vaccination use of other antigens, better estimates of quantifying vaccine demand and its resulting ®nancing could be made. As vaccine production costs are scale dependent [10], better estimates of global demand forecasts would allow manufacturers to plan for capacity that may ultimately result in lower costs [11]. This possibility is especially relevant with the development of newer vaccines that have the potential of serving global populations. Timely forecasts may have great implications on production planning with the possibility of more aordable vaccine at the local level. For example, the potential for practical utility of the model has been demonstrated by applying the HB model coecients to the Hib database and testing its predictive value. The model predictions showed strong
2229
agreement between the estimated probability of Hib use with current vaccine adoption. By ranking countries by their estimated probability to uptake vaccine, countries might be appropriately targeted for marketing purposes and estimates of vaccine demand can be forecast. This may be of interest to manufacturers and public sector purchasing agencies such as UNICEF. In particular, the model allows one to conduct a ``what if'' analysis. For example, how would an individual country's probability and consequently global demand change with dierent vaccine costs. Furthermore the models may be modi®ed to utilise existing databases for vaccines such as yellow fever, rotavirus and pneumococcus to help estimate vaccine needs. 4.2. Parameters to in¯uence As vaccines are frequently under-utilised, the model highlights the parameters that might be modi®ed to increase the probability of uptake. Of the most sensitive factors (vaccine cost, per-capita GDP, infrastructure, perception of disease burden and population size), a few might be in¯uenced to increase the probability of vaccine uptake. For example, development agencies will have limited in¯uence on per-capita GDP and population size in the short run, but they can work with manufacturers to in¯uence the vaccine cost to the country (either through ®nancial support for vaccine purchase or manufacturers lowering vaccine prices). A quantitative estimate of how much price support is required for a particular country to move it into a high probability adoption status may be calculated from the models. In addition, vaccination infrastructure, as measured by DTP coverage data, could be strengthened to increase the probability of new vaccine uptake. Finally, disease burden assessments can be performed to increase the perception of disease in speci®c locales that may represent large geographic or socio±economic groups. Any change of these parameters could increase the probability of vaccine use. For example, one could calculate how much ®nancial support would be required for a particular country to increase their probability of Hib vaccine uptake from 25 to 75%. Likewise, one could identify the minimum level of disease burden that must be demonstrated to signi®cantly increase the probability of vaccine uptake. 4.3. Limitations These results are limited by the degree of data resolution. The data did include sub-national estimates of parameters [e.g. public/private sector, high-risk populations (geographic, ethnic)] and oer results as probability estimates in a particular time period. Other
2230
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
possible associations not incorporated in the model are the regional in¯uence of WHO oces, speci®c industry marketing practices countries, and changes over time. However, the model may be further re®ned to re¯ect the impact of time if further time series data could be collected. Speci®cally, a price elasticity curve (price of vaccine to countries over time vs cumulative number of countries adopting vaccine and doses sold would be useful. This work is currently in progress. Although vaccination status was strongly associated with the factors we identi®ed, an association does not prove cause and eect. Nevertheless, the associations are strong, results plausible, and may help provide a basis for planning vaccine production and for global and regional vaccine policy.
Acknowledgements We would like to thank Roy Widdus, Julie Milstien, and Steve Landry for their commentary and review; Laura McCann and Diana Chang for their editorial review.
References [1] World Health Organization, United Nations Children's Fund. State of the world's vaccines and immunization. Geneva: World Health Organization, 1996. [2] World Health Assembly. In: Immunization and vaccine quality. Forty-®fth World Health Assembly, Geneva 4±14 May 1992, 1992 Resolution WHA45.17. [3] Global Programme for Vaccines Immunization. The WHO position paper on Haemophilus in¯uenzae type b conjugate vaccines. Weekly Epidemiological Record 1998;73:64±8. [4] Miller M, McCann L. Policy analysis of the use of hepatitis b, Haemophilus in¯uenzae type b-, Streptococcus pneumoniae-conjugate and rotavirus vaccines in national immunization schedules. Health Economics 2000 (in press). [5] United Nations Population Division. World Population Prospects The 1996 revision. New York: United Nations Publications, 1997. [6] World Bank. World Development Report 1997: the State in a changing world. Washington DC: World Bank, 1997. [7] World Bank. World Development Report 1993: investing in health. New York: Oxford University Press, 1993. [8] Colton T. In: Statistics in Medicine. Boston: Little Brown and Company, 1974. p. 221±3. [9] Lemeshow S, LeGall SR. Modeling the severity of illness of ICU patients: a systems update. JAMA 1994;272:1049±55. [10] Mercer Management Consulting. Report on the United States vaccine industry. Geneva: Children's Vaccine Initiative, 1995. [11] Batson A. Win±win interactions between the public and private sectors. Nature Medicine 1998;4:487±91.
www.elsevier.com/locate/vaccine
A model to estimate the probability of hepatitis B- and Haemophilus in¯uenzae type b-vaccine uptake into national vaccination programs Mark A. Miller a,*, W. Dana Flanders b a
Children's Vaccine Initiative Secretariat, c/o World Health Organisation, 20 Avenue Appia, 1211, Geneva, Switzerland b Emory University, Atlanta, GA, USA Received 16 March 1999; received in revised form 17 November 1999; accepted 7 December 1999
Abstract Most countries have been slow to adopt new vaccines into national vaccination schedules, despite recommendations from global multi-lateral agencies. Characteristics of countries that have adopted hepatitis B (HB) vaccine were analysed and used to formulate a logistic regression model. The model was applied to country-speci®c data to predict HB and Haemophilus in¯uenzae type b (Hib) vaccine uptake. The greatest predictors of HB uptake were coverage rates of other vaccines, vaccine cost relative to the economy, and perceived disease burden. The logistic regression model's probability estimate of vaccine uptake agreed well with observed data for HB and Hib, (c-statistic 85 and 82%, respectively). Application of this model to other antigens may aid in predicting potential national markets to better plan new vaccine supply and demand. 7 2000 Published by Elsevier Science Ltd. All rights reserved. Keywords: Hepatitis B; Haemophilus in¯uenzae type b; Vaccines; Economics
1. Introduction After the success of smallpox eradication, the World Health Organisation's (WHO) Expanded Programme on Immunisation (EPI) has helped to introduce six basic vaccines, diphtheria±tetanus±pertussis (DTP), Bacille Calmette±GueÂrin, measles, and poliomyelitis, into national vaccination programs throughout the world. In 1988, the EPI recommended yellow fever vaccine to be added to national immunisation schedules for use in countries with endemic disease [1]. The World Health Assembly recommended hepatitis B (HB) vaccine for universal childhood vaccination in 1992 [2]. Most recently, the WHO recommended Hae* Corresponding author. After February 1: Fogarty International Center, National Institutes of Health, Bethesda, MD 20892, USA. Tel.: +1-301-496-1415; fax: +1-301-496-2173. E-mail address: [email protected] (M.A. Miller).
mophilus in¯uenzae type b (Hib) conjugate vaccines for use in routine immunisation programs in countries with a demonstrated disease burden, ``as appropriate to national capacities and priorities'' [3]. Despite these recommendations, many countries have been slow to adopt these vaccines in routine immunisation schedules. Yellow fever vaccine has been available since 1937, yet fewer than 13 of the 45 countries where yellow fever is endemic report routine use of this vaccine (EPI, unpublished data). HB vaccine was ®rst licensed in 1981, however, less than one-half of the world's children born each year receives this vaccine in routine schedules [4]. Despite the availability of Hib conjugate vaccines for almost 10 years, their use is limited to less than 20% of the global birth cohort (J. Wenger, WHO, unpublished data). Numerous factors have been cited for the slow uptake of new vaccines. These include vaccine cost,
0264-410X/00/$ - see front matter 7 2000 Published by Elsevier Science Ltd. All rights reserved. PII: S 0 2 6 4 - 4 1 0 X ( 9 9 ) 0 0 5 6 3 - 0
2224
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
poor infrastructure to deliver vaccines and lack of adequate data on disease burden. However, there has been no systematic quanti®cation of the importance of these factors. With the development of new eective vaccines against enteric disease (rotavirus, cholera, typhoid), pneumonia (pneumococcal conjugate) and meningitis (meningococcal conjugate), estimating potential vaccine demand and quantifying the magnitude of the barriers for their introduction is becoming increasingly important. Using HB vaccine uptake as a model, we quantify the factors that are associated with vaccine uptake into national immunisation schedules and apply a model to estimate the probability of HB and Hib vaccine adoption. 2. Methods The goal of these analyses is to identify and assess factors associated with the adoption of HB vaccines into routine national vaccination programs. Characteristics of countries that have adopted HB vaccine were compared to those countries that have not. Logistic regression models were developed based on country speci®c epidemiologic and economic factors. These factors were then applied to data speci®c to Hib disease and vaccine to assess the model's predictive ability of Hib vaccine uptake.
Table 1 List of variables studied Variable Demographics 1996 Surviving infants and total population Geopolitical region WHO Region Economic 1996 Per-capita GDP World Bank income group Program factors Vaccine cost (per dose) DTP vaccine coverage Health characteristics Life expectancy at birth HB surface antigen prevalence Deaths in current birth cohort due to HB Years of life lost per birth Unit case treatment cost Integrated factors Vaccine program cost Deaths prevented per 1000 infants Life-years saved per 1000 infants Treatment costs prevented per 1000 infants a
UNPD±United Nations Population Division. WB±World Bank. c WHO±World Health Organisation. d CVI±Children's Vaccine Inititative. b
Data source UNPDa WBb WHOc WB WB CVId WHO WHO WHO CVI CVI CVI CVI CVI CVI CVI
The HB and Hib models were based upon countryspeci®c epidemiologic and economic data previously described [4]. Of the 215 countries or territories that report data to the WHO, 36 were excluded due to data limitations. Of those excluded, all have populations less than 150,000 and most are island nations. Countries were characterised as having no routine scheduled HB vaccination program (either infant/childhood or adolescent), planning a HB vaccination program or current HB vaccination program as of October, 1998. 2.1. Data collection Published data pertaining to HB vaccination programs, which include country speci®c demographics, economic data and data speci®c to their national vaccination program, served as variables for the model. Speci®c variables and data sources are speci®ed in Table 1. Factors considered as potential predictors include descriptive data of each country's demographic features (total population, birth cohort) [5] and economy as de®ned by per-capita gross domestic product (GDP) [6] and geo-political category as de®ned by the World Bank [7]. Program factors assessed for each country included the unit cost of vaccination, including vaccine and administration; and the coverage rate for DTP vaccine. Measures of health characteristics of each country assessed included: the prevalence of HB surface antigen (hbsag), an indicator of hepatitis B endemicity, and the estimated annual incidence of deaths and years of life lost per infant (YLL) due to the speci®c vaccine preventable disease. The methodology for the calculation of mortality and YLL are summarized in a previous study [4]. Integration of these variables was also assessed. For example, various health outcomes, such as the reduction in incidence of death and years of life lost through HB vaccination were determined by integrating disease burden and coverage rates. In addition, treatment costs prevented is based on integrating disease burden, coverage and economic factors. Country speci®c determinations were adjusted for the size of the birth cohort. The methodology for the derivation of these characteristics has been previously described [4]. 2.2. Analyses In exploratory analyses, the mean and standard deviation for each potential predictor variable were compared between countries that adopted HB vaccine into its national vaccination schedule with the corresponding measures among countries that had not adopted such a policy. We use the Wilcoxon rank sum test [8]
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
to test for dierences between the distribution of continuous variables in countries adopting a vaccination policy versus other countries. Logistic regression was used to study further predictors of each country's decision to adopt a national vaccination program for HB. Each variable was grouped into quartiles (or other natural groupings) to assess the general pattern of the relationship between vaccination and the variable. If the exploratory analysis was consistent with a monotonic dose±response, the variable was used as a continuous variable in subsequent logistic regression analyses. We identi®ed a predictive model for adoption of vaccination by ®rst using a priori considerations and the exploratory analyses to select key variables. These variables were further assessed in multivariate analyses for their predictive ability while, simultaneously controlling for the other variables. To illustrate the association between each variable and vaccination status, we calculated odds ratios (OR) as follows: the OR represents the odds that a country at the mean of the highest quartile would adopt vaccine relative to the corresponding odds for a country in the lowest quartile (OR=exp[b(x4ÿx1)]; where b=estimated regression coecient, x4=the mean value of the independent variable for countries in the highest quartile of that variable and x1=the corresponding mean for the lowest quartile).
2225
2.3. Regression diagnostics We screened for potential collinearity by calculating correlation coecients. The logistic regression model was then applied to the country-speci®c data for HB and Hib to determine each country's probability of vaccine uptake. These estimated probabilities were compared to the observed proportion of countries that had committed to adopt HB into their routine infant immunization schedule as of October 1998. We calculated the area under the receiver operating characteristic curve (c-statistic) to indicate the discriminatory ability of the model to distinguish countries that vaccinated from those that did not [9]. We used the Hosmer±Lemeshow goodness of ®t test to assess model calibration [9]. 3. Results Of the 179 countries studied, 75 (42%) have adopted HB into their routine vaccination schedule as of October 1998, (Fig. 1). The percentage of countries using HB vaccine varied regionally from 11% of countries in the World Health Organisation's (WHO) African region to 79% in the Western Paci®c region (Table 2). The percentage using HB also varied by income group
Fig. 1. Model prediction of the probability of HB vaccine uptake and countries with current routine use. The estimate of the probability of adoption, determined by the logistic regression, is proportional to the degree of shading for each country. The circles indicate countries that are currently using HB vaccine.
2226
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
Table 2 Frequency of adoption of vaccination policy, by geopolitical and economic characteristics Variable WHO Region African American Eastern Mediterranean European Southeast Asian Western Paci®c World Bank Income Group Classi®cationa Low (GNP R $785) Lower middle ($785 < GNP R $3125) Upper middle ($3125 < GNP R $9655) High (GNP > $9655) World Bank Classi®cation China Established market economies Former socialist economies of Europe India Latin America and Carribean Middle-east Crescent Sub-Sahara Africa Other Asia and Paci®c Islands a
Total number of countries
Number vaccinating (percentage of total in parentheses)
44 36 23 47 10 19
5 (11.4%) 9 (25.0%) 16 (69.6%) 25 (53.2%) 5 (50.0%) 15 (79.0%)
48 49 44 38
4 (8.3%) 24 (49.0%) 19 (43.2%) 28 (73.7%)
1 25 23 1 34 26 43 26
1 (100.0%) 17 (68.0%) 9 (39.1%) 0 (0.0%) 7 (20.6%) 18 (69.2%) 5 (11.6%) 18 (69.2%)
Income groups are classi®ed based on per capita gross national product (GNP) as measured in US dollars.
and geopolitical characterisation, as de®ned by the World Bank. High-income countries were more likely to adopt HB into routine schedules then low-income countries (73.7 vs 8.3%, OR 8.8, p < 0.0001). Several continuous factors were also associated with HB vaccine uptake (Table 3). Per-capita GDP ($8475
vs 3236) and vaccine cost ($3.96 vs 1.65 per dose) were signi®cantly higher in those countries that routinely vaccinate against HB than those that do not. However, when the vaccine cost was adjusted as a percentage of the country's GDP, the reverse is true (0.17 vs 0.40). Countries that adopted HB had a mean DTP coverage
Table 3 Means of selected factors, by countries that have adopted HB vaccinea Variable Demographic Surviving infants (thousands) Total population (millions) Economic Per-capita GDP (USD) Program factors Vaccine cost per dose (USD) Vaccine cost as percentage of per capita GDP DTP Coverage (percentage) Health characteristics HB surface antigen prevalence (percent) Estimated deaths (per 1000 infants) Estimated years of life lost (per 1000 infants) Unit case treatment cost (USD) Integrated factors Vaccine program cost (per birth) (USD) Deaths prevented by vaccination (per 1000 infants) Estimated life-years saved (per 1000 infants) Estimated treatment costs prevented (per 1000 infants) (USD) a
Mean (SD) in countries not vaccinating
Mean (SD) in countries vaccinating
p-value for dierence in means (Wilcoxon)
727 (2458) 27 (96)
745 (2507) 40 (146)
NS NS
3236 (6481)
8475 (9386)
< 0.0001
1.65 (3.00) 0.40 (0.37) 79.5 (19.8)
3.96 (4.48) 0.17 (0.11) 93.8 (5.8)
< 0.0001 < 0.0001 < 0.0001
6.4 (5.1) 13 (10) 135 (131) 1595 (2790)
5.8 (4.9) 12 (10) 270 (218) 3857 (3825)
NS NS < 0.0001 < 0.0001
9.42 (17.96) 9 (7)
23.49 (26.18) 10 (9)
< 0.0001 NS
102 (114) 230 (496)
239 (193) 1381 (2406)
< 0.0001 < 0.0001
Variables in bold are signi®cantly dierent at the p < 0.0001 level.
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
2227
Table 4 Univariate associations of selected factors with vaccination status, based on logistic modela Variable Economic Per Capita GDP Program factors Vaccine cost (per dose) Vaccine cost/per-capita GDPc DTP coverage Health characteristics HB surface antigen prevalence Deaths (per 1000 infants) Years Life Lost (per 1000 infants) Unit treatment cost Integrated factors Deaths prevented Years life saved Treatment cost prevented
Odds ratiob
95% Con®dence intervals
4.4
2.1
9.4
4.1 39.7 55.1
1.9 6.5 10.4
8.6 241.2 292.6
0.7 0.7 6.9 5.0
0.4 0.8 2.9 2.3
1.5 1.5 16.1 11.0
1.6 9.6 127.4
0.8 3.8 15.5
3.4 23.9 1043.6
a
Bold type indicates signi®cant variables. Odds ratios represent highest quartile relative to lowest quartile in each category. c Lowest quartile relative to highest quartile was compared to yield an OR greater than 1 for comparisons to other variables.
b
greater than countries that have not adopted HB vaccine (93.8 vs 79.5%). Carriage prevalence, as measured by HB surface antigen, did not dier substantially between the two groups, nor did the estimated deaths that would occur in the absence of vaccination. However, life expectancy in a country, when integrated with HB mortality to estimate an incidence of years of life lost due to HB, was strongly associated with adoption of HB vaccine. In addition, countries that spent more on HB treatment per capita were more likely to have adopted HB vaccine. When these factors were combined, countries that have adopted HB vaccine, on average, were paying more per capita for their HB vaccine program ($23.49 vs 9.42) but were saving more life years (239 vs 102) and more in HB treatment costs ($1381 vs 230) per 1000 infants.
larly strongly associated with vaccine adoption (OR 127.4, 95% CI 15.5±1043.6). We identi®ed three characteristics of countries that were strongly associated with vaccine adoption, even after controlling for the other factors. Vaccine cost as a percentage of per capita GDP, DTP coverage, and YLL per 1000 infants were each strongly associated with increased likelihood of adoption of a vaccination policy. Together these three factors strongly predict a country's policy, with a c-statistic of 85%, and provide a good ®t to the observed data as determined by the Hosmer±Lemeshow test. Based on the model coecients and country speci®c data, a map of the probability of hepatitis B uptake was constructed (Fig. 1). Correlation with the outcome is illustrated since countries with a higher predicted probability of vaccine usage tended to actually be using vaccine as of October 1998.
3.1. Logistic regression Results of logistic regression also showed that several factors were strongly associated with adoption of a vaccination program (Table 4). Vaccine adoption was strongly associated with coverage rates of DTP vaccine (OR 55.1 95% CI 10.4±292.6), vaccine cost relative to per-capita GDP2 (OR 39.7, 95% CI 6.5± 241.2) and estimated YLL per infant in the current birth cohort (OR 6.9 95% CI 2.9±16.1). Treatment cost saved through vaccination (which integrates the unit cost of treatment and coverage rates) was particu2 Lowest quartile relative to highest quartile to yield a OR greater than 1 for comparison to other variables.
3.2. Hib The logistic regression model coecients derived from the HB data were applied to the country speci®c data in the Hib database (Hib vaccine cost, GDP, estimated YLL due to Hib, and vaccine coverage rates) to estimate probabilities of Hib vaccine uptake (Fig. 2). The map illustrates agreement of the probability estimates with the observed data. The c-statistic was 82%, indicating that the model could distinguish between countries that adopted vaccines from those that did not.
2228
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
Fig. 2. Model prediction of the probability of Hib vaccine uptake and countries with current routine use. The estimate of the probability of adoption, determined by application of the parameter estimates from the HB model to the Hib data, is proportional to the degree of shading for each country. The circles indicate countries that are currently using Hib vaccine.
3.3. Estimated demand of vaccine doses
4. Discussion
Based on the models' estimate of a country's probability of adopting HB or Hib vaccines, we have calculated the number of required vaccine doses in each category (Table 5). For example, assuming equivalent rates of vaccine use to DTP coverage, 21 million doses would be required annually in countries that have a >75% probability for vaccine uptake. An additional 16 million doses would be required for countries with a >50 to 75% probability of Hib uptake.
Our results highlight the relative importance of the national commitment to the vaccination programs (coverage rates of DTP) and economic indicators relative to disease burden estimates associated with vaccine uptake. Surprisingly, carriage prevalence and number of deaths due to HB did not correlate with vaccine adoption, however, when years of life lost were incorporated, a disease burden correlate was found. We speculate that countries where average life expectancy
Table 5 Distribution of the number of HB and Hib doses required per annum based on the model's estimate probability of vaccine adoptiona Total doses of vaccine (millions) required by countries in each category Estimated probability of vaccine uptake
HB
Hib
0±25% > 25±50% > 50±75% > 75% Total
151 19 54 136 361
333 4 16 21 374
a
Dierences in the total doses result from omission of smaller countries in the HB dataset.
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
is less than 50 are less likely to recognise deaths due to HB, which occur at mid- to late-stage of life, to be a major public health problem relative to those countries whose populations have longer life expectancies and, consequently, years of life lost. Interestingly, the combined variable of treatment costs saved through vaccination, which incorporates coverage, treatment expenditures and disease burden, had the strongest association. The model predicted quite well the probability that a country had adopted HB and Hib vaccines, as re¯ected in the ®gures. For HB, the model did not correctly predict adoption for a few countries, almost exclusively in Latin America, Central Asia and Scandinavia. The low use in Scandinavia may be attributed to the relatively low incidence of disease in the native population and to countries which have an alternative HB vaccination policy of targeting high-risk groups (i.e. infant immunisation in immigrants). Although all ®ve Central Asian nations of the former Soviet Union were estimated to have greater than 50% probability of using HB, only two countries had done so. The recent decentralisation of government health services may have negatively impacted on the introduction of new vaccines. Finally, in Latin America, much of the vaccine policy and purchase is co-ordinated by the Pan American Health Organization (PAHO) which may have impacted on the adoption of vaccine into routine schedules. Some countries in this region have introduced HB vaccine to target high-risk populations in certain geographical regions (e.g. populations in the Amazon). In 1999, due to centralised purchasing, numerous countries in the PAHO region will adopt HB and Hib vaccine, further supporting the model's probability estimates. 4.1. Utility of model Ð market forecasts and demand estimates By categorising the countries by their probability of vaccine uptake and linking it with data of national vaccination use of other antigens, better estimates of quantifying vaccine demand and its resulting ®nancing could be made. As vaccine production costs are scale dependent [10], better estimates of global demand forecasts would allow manufacturers to plan for capacity that may ultimately result in lower costs [11]. This possibility is especially relevant with the development of newer vaccines that have the potential of serving global populations. Timely forecasts may have great implications on production planning with the possibility of more aordable vaccine at the local level. For example, the potential for practical utility of the model has been demonstrated by applying the HB model coecients to the Hib database and testing its predictive value. The model predictions showed strong
2229
agreement between the estimated probability of Hib use with current vaccine adoption. By ranking countries by their estimated probability to uptake vaccine, countries might be appropriately targeted for marketing purposes and estimates of vaccine demand can be forecast. This may be of interest to manufacturers and public sector purchasing agencies such as UNICEF. In particular, the model allows one to conduct a ``what if'' analysis. For example, how would an individual country's probability and consequently global demand change with dierent vaccine costs. Furthermore the models may be modi®ed to utilise existing databases for vaccines such as yellow fever, rotavirus and pneumococcus to help estimate vaccine needs. 4.2. Parameters to in¯uence As vaccines are frequently under-utilised, the model highlights the parameters that might be modi®ed to increase the probability of uptake. Of the most sensitive factors (vaccine cost, per-capita GDP, infrastructure, perception of disease burden and population size), a few might be in¯uenced to increase the probability of vaccine uptake. For example, development agencies will have limited in¯uence on per-capita GDP and population size in the short run, but they can work with manufacturers to in¯uence the vaccine cost to the country (either through ®nancial support for vaccine purchase or manufacturers lowering vaccine prices). A quantitative estimate of how much price support is required for a particular country to move it into a high probability adoption status may be calculated from the models. In addition, vaccination infrastructure, as measured by DTP coverage data, could be strengthened to increase the probability of new vaccine uptake. Finally, disease burden assessments can be performed to increase the perception of disease in speci®c locales that may represent large geographic or socio±economic groups. Any change of these parameters could increase the probability of vaccine use. For example, one could calculate how much ®nancial support would be required for a particular country to increase their probability of Hib vaccine uptake from 25 to 75%. Likewise, one could identify the minimum level of disease burden that must be demonstrated to signi®cantly increase the probability of vaccine uptake. 4.3. Limitations These results are limited by the degree of data resolution. The data did include sub-national estimates of parameters [e.g. public/private sector, high-risk populations (geographic, ethnic)] and oer results as probability estimates in a particular time period. Other
2230
M.A. Miller, W.D. Flanders / Vaccine 18 (2000) 2223±2230
possible associations not incorporated in the model are the regional in¯uence of WHO oces, speci®c industry marketing practices countries, and changes over time. However, the model may be further re®ned to re¯ect the impact of time if further time series data could be collected. Speci®cally, a price elasticity curve (price of vaccine to countries over time vs cumulative number of countries adopting vaccine and doses sold would be useful. This work is currently in progress. Although vaccination status was strongly associated with the factors we identi®ed, an association does not prove cause and eect. Nevertheless, the associations are strong, results plausible, and may help provide a basis for planning vaccine production and for global and regional vaccine policy.
Acknowledgements We would like to thank Roy Widdus, Julie Milstien, and Steve Landry for their commentary and review; Laura McCann and Diana Chang for their editorial review.
References [1] World Health Organization, United Nations Children's Fund. State of the world's vaccines and immunization. Geneva: World Health Organization, 1996. [2] World Health Assembly. In: Immunization and vaccine quality. Forty-®fth World Health Assembly, Geneva 4±14 May 1992, 1992 Resolution WHA45.17. [3] Global Programme for Vaccines Immunization. The WHO position paper on Haemophilus in¯uenzae type b conjugate vaccines. Weekly Epidemiological Record 1998;73:64±8. [4] Miller M, McCann L. Policy analysis of the use of hepatitis b, Haemophilus in¯uenzae type b-, Streptococcus pneumoniae-conjugate and rotavirus vaccines in national immunization schedules. Health Economics 2000 (in press). [5] United Nations Population Division. World Population Prospects The 1996 revision. New York: United Nations Publications, 1997. [6] World Bank. World Development Report 1997: the State in a changing world. Washington DC: World Bank, 1997. [7] World Bank. World Development Report 1993: investing in health. New York: Oxford University Press, 1993. [8] Colton T. In: Statistics in Medicine. Boston: Little Brown and Company, 1974. p. 221±3. [9] Lemeshow S, LeGall SR. Modeling the severity of illness of ICU patients: a systems update. JAMA 1994;272:1049±55. [10] Mercer Management Consulting. Report on the United States vaccine industry. Geneva: Children's Vaccine Initiative, 1995. [11] Batson A. Win±win interactions between the public and private sectors. Nature Medicine 1998;4:487±91.