Although pertussis is a mandatory notifiable disease regardless of age in all countries in this study [30,31,32,33,34], many cases in adults are not captured by surveillance systems. This issue of underestimation is also recognized in other countries. For example, a community cohort study in healthy people in South Africa estimated the incidence of pertussis at 0.21 per 100 person-weeks, higher than the previously reported mean annual incidence of 17 per 100,000 population in patients hospitalized for pneumonia [35]. To our knowledge, this is the first study in Latin America to estimate the underestimation rate for pertussis in adults aged 50 years or over based on a probabilistic model using data from literature and national surveillance systems. Our results indicated that the number of pertussis cases in these countries is approximately 104–114 times higher than the number reported to national surveillance systems. The 90% limits around the estimated median values were broad, with a lower limit of 39 to 41 and an upper limit of 419 to 482, reflecting some uncertainty in some of the input values for the pertussis parameters in the model. For pertussis hospitalizations, the median estimated multipliers were smaller and had narrower 90% limits (in Brazil, median 2.3, lower and upper 90% limits 1.8, 3.3; in Chile, median 2.4, lower and upper 90% limits 1.8, 3.2). This is because the data inputs used for the proportion seeking healthcare and the proportions with a specimen collected and sent for testing were higher and less variable in patients hospitalized for pertussis disease, compared with non-hospitalized cases. The median multiplier estimates were similar for all five countries, because the same pertussis input data were used across all countries, and the variability in the multipliers was derived from the probabilistic model. If the study had used different input data values for each country, it is likely that the multipliers would also have differed between the countries. We were not able to use country-specific input data because of a lack of available information. The limited real-world surveillance data available for pertussis cases in adults in the selected countries might suggest differences in the multipliers between the countries. However, considering the scarcity of parameter data for pertussis in adults, we standardized the input data in the present simplistic model, resulting in little difference between the resulting country multipliers. The results aim to serve as baseline data that can be used in sensitivity analyses in further modelling studies. Even with the small multiplier range, our results provide a baseline estimate for specific country pertussis incidence, adjusted to country-specific population and surveillance data stratified by age group.

Our results are broadly consistent with previous studies in Europe and Israel that have also estimated pertussis reporting rates to be hundreds of times lower than the number of infections estimated from serological data. In Israel, the incidence of pertussis infection in 2000, estimated from serology, was 2448 per 100,000 population aged 3 years or over, compared with 5.6 per 100,000 for reported pertussis cases in the same year, indicating an incidence of infection approximately 400 times higher than the incidence of notified clinical pertussis cases [36]. In the Netherlands, pertussis incidence estimated by serology was 6.6% per year for people aged 3–79 years, approximately 685 times higher than the 0.01% annual incidence of notified cases [37]. Our estimated median case numbers of approximately 100 times higher than reported data are somewhat lower than the results from these studies, but the 90% upper limit of our estimates (419 to 482) is similar to the value reported in the Israeli study [36]. Our results are also broadly consistent with a study in the USA that estimated undiagnosed pertussis cases in adolescents and adults with reported acute respiratory disease using clinician notes in an electronic health record database, using a machine learning algorithm [38]. This US study found that accounting for undiagnosed pertussis episodes increased the estimated pertussis incidence by 110-fold on average [38]. This estimate is very similar to our median estimated multiplier of 104–114.

There is a consensus that reported pertussis cases are likely to be considerably lower than the real number of pertussis cases [39]. Underreporting is a particular problem in older age groups, as the only symptom in adults and adolescents may be prolonged cough, for which patients may present late or not at all, and which may be misdiagnosed if clinicians perceive pertussis as a childhood disease [39]. In Italy, the estimated incidence of pertussis based on seroprevalence data in the population aged ≥ 15 years was 3464 times higher than the rate of notified pertussis cases in 2018, while in the population aged 6–14 years the estimated incidence from seroprevalence was 141 times the notification rate [40]. A study in the Netherlands reported a remarkable difference in the age distribution for notified pertussis cases compared with infection rates estimated from serological data [37]. The incidence of notified cases was highest in infants (77.2 per 100,000 in the first year of life) and young children (87.4 per 100,000 in children aged 3–4 years and 63.1 per 100,000 in children aged 5–9 years), decreasing sharply to around 1–3 per 100,000 in adults aged 20 years or over. By contrast, the estimated incidence was lowest in children aged 3–4 years (3299 per 100,000), reached a peak in young adults aged 20–24 years (10,831 per 100,000), and fluctuated around 6,500 per 100,000 in adults aged 25–54 years and around 4000 per 100,000 in adults aged 55 years or over [37].

The number of pertussis cases per 100,000 population estimated in this study varied between countries, with the highest estimates in Chile. This reflects the higher number of pertussis surveillance reports in Chile, which results in a larger number of cases as the multiplier is similar as a result of a lack of country-specific data for the pertussis inputs in the model. The present study was not intended to make comparisons between the countries. It illustrates the potential magnitude of underestimation of pertussis, and there is considerable uncertainty in the estimates, as indicated by the broad 90% limits around the median values. In a study comparing the number of isolates of invasive pneumococcal disease reported to surveillance with the expected number of cases based on regional disease incidence data, Chile had the highest rate of reporting (43–83%) of the six Latin American countries in the study [41]. Other researchers have reported wide variations in pertussis prevalence between countries; in a study in 12 European countries conducted in 2007–2010, the percentage of adult patients presenting to primary care with acute cough who were found to have B. pertussis ranged from 0% in Italy to 6.2% in Sweden [42]. A review of the literature reported rates of pertussis underreporting varying from 5 to 50% in Brazil, Colombia, Spain, and the USA, and in the Netherlands a multiplier factor of 3 for infants and 200 for adults [43]. However, comparison of pertussis incidence between countries is difficult because of factors such as differences in case definitions, diagnosis methods, and reporting systems [39]. Open-source data may provide useful information on pertussis outbreaks globally, especially in countries that lack national surveillance systems [44]. Most studies of the cost-effectiveness of pertussis vaccination in adolescents and adults attempt to take account of underreporting of pertussis cases in these age groups. For example, a review of 27 economic evaluations of vaccination strategies for tetanus–diphtheria–acellular pertussis vaccine (Tdap) in a range of countries found that 18 of the 20 studies that considered adult or adolescent pertussis incidence used some strategy to correct for pertussis underreporting [45].

In most of the countries in the study, the highest number of pertussis cases per 100,000 population was in the oldest age group, aged 90 years or over. The data in Table 1 indicate that the five countries in the study have slightly different population structures, with the percentage of people in the group aged 50–59 years highest in Brazil (45%), Mexico (47%), and Peru (45%). This may indicate a potential future increase in the burden of pertussis in these countries as this group reaches older ages.

The present study has a number of strengths. It used an established model developed by the Centers for Disease Control and Prevention which has previously been applied to influenza [16] and hepatitis [17]. Furthermore, it provides a simplified way to visualize and quantify pertussis underestimation in older adults in Latin America. This should help to increase awareness of the health burden of pertussis in adults among health officials and clinicians, helping to support surveillance and prevention activities.

Nevertheless, the study has a number of limitations. Data for hospitalizations were available for only two countries, Brazil and Chile, and therefore the estimated multiplier for hospitalizations was based only on these two countries as there were no others to construct the probabilistic model. Further data on hospitalizations from other countries would improve the analysis, if such data become available in the future. Surveillance data from Argentina were also limited, with a total of two hospitalizations (one from 2017 and one from 2018) and zero cases reported across the whole study period. It is not feasible to run the model with a data input of zero, so the model for Argentina was run using the two hospitalized cases as a minimum proxy for ambulatory cases. With these very small numbers, it is possible that even the predicted cases could have a degree of underestimation. The study collected data from 1 January 2017 to 31 December 2021, and this included the period of the COVID-19 pandemic, which could have affected data collection. We constructed the model with the ability to split the analysis into two separate periods, 1 January 2017 to 31 December 2019 and 1 January 2020 to 31 December 2021, so that the COVID-19 pandemic period could be excluded if necessary. However, there was little difference in surveillance results between the two periods, so results are presented across the whole study duration. The analysis involved several assumptions, and data for parameter estimates were sparse, indirect and sometimes limited in geographical scope. Few studies have assessed underestimation in adults with pertussis. In Latin America, we identified only one published source with an estimate of the frequency of healthcare-seeking among adults with pertussis [27], and one reporting results from a Delphi panel [29]. Both these publications were from Brazil, reflecting its position as the largest country in the region. The issues with pertussis diagnosis and testing in adults are similar across all countries in Latin America, so it is likely that data from Brazil will be applicable to other countries in the region. For other inputs, we had to use data from countries outside Latin America, which may not be representative of the countries in the analysis. To account for some of this uncertainty, we used a probabilistic model with a range of values for each parameter. In addition, the case definition of pertussis may have varied between the studies from which parameter values were obtained, and since adults with milder illness may be less likely to seek care or be tested, they may not be fully captured in these estimates. Furthermore, we did not evaluate variability or uncertainty in the model parameters using sensitivity analysis. If more accurate and detailed parameter data become available in the future from surveys of health-seeking behavior, testing practices and policies, this would help to improve the estimates in this study. In the meantime, the estimates presented here should help to support decision-making for the management of pertussis in older adults in Latin America.