Introduction

Lung cancer has progressed to be one of the most frequent and deadly variations of cancer worldwide, even in low-middle-income countries (LMICs) [1]. Despite advances in diagnosing and treating this disease, lung cancer prevalence and mortality have increased. According to the last report of The Global Cancer Observatory (GLOBOCAN), this type of cancer represents the leading cause of mortality in cancer, with 23 deaths per 100,000 being reported globally [2, 3]. The low prevalence rate in LMIC is a concern because 80% of smokers live in these countries [4].

Health outcomes are worse in Latin American countries than in other regions because of their fragmented and underbudgeted health systems [5]. For optimal cancer care, LMICs must demonstrate significant prevention and early diagnosis strategies, as well as high quality measures, in medication, surgery and radiation delivery backed by patient support. However, health systems in LMICs are precarious, and most people with this condition are diagnosed in late stages if their cancer occurs, thus causing higher mortality rates [6].

Lung cancer in Colombia has an important impact, as it represents the fourth place in mortality among men and the fifth place among women [7]. Furthermore, a demographic change in the population has been described, which indicates an increase in individuals aged over 60-years-old [8]. It has also been described that lung cancer risk increases over time, especially in elderly people over 65-years-old [9, 10]. All of these combined factors can cause an increase in prevalence rates in some regions of the country due to the high prevalence of smokers or mining activities [11, 12].

The Colombian health system covers 99% of Colombian citizens [13]. It has a centralized fund controlled by the government that gathers payroll taxes and contributes an additional amount of money to ensure health care for the population, including the poorest individuals of this population. At the level of the insurers, there are public and private health institutions (known as Empresas Promotoras de Salud [EPS]) that act as intermediate payers between the centralized fund and health service providers (known as Instituciones Prestadoras de Salud [IPS]). These IPSs compete to be included in each EPS service network. Moreover, the insured population can be classified into a group that contributes to the health system via payroll taxes (known as the contributory regime) and a group funded by the national government (known as the subsidized regime) [14, 15].

This study aimed to develop an electronic algorithm to identify lung cancer-prevalent patients over 20-years-old in Colombia by using official databases, as well as to estimate prevalence rates of lung cancer by age, sex, and geographic region. This information could provide valuable information to Colombian policymakers, clinicians and the world to improve knowledge about lung cancer in LMICs.

Methods

Results

After testing 16 algorithms, two electronic algorithms were selected, as they had the closest prevalence rates to those rates reported by aggregated official sources (GLOBOCAN [18], National Cancer Institute [19] in Colombia and Cuenta de Alto Costo-Colombia [20]). These comparisons were evaluated by comparing point estimates provided the mentioned sources to the confidence intervals calculated using bootstrapping with 1 000 subsamples as shown in Tables 1 and 2. The first selected algorithm was defined as the presence of ICD-10 lung cancer codes for four months or more (the sensitive algorithm), and the second selected algorithm was defined as the same criteria from the first algorithm and the presence of at least one oncological procedure (the specific algorithm).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Comparison between the prevalence rates identified in this study versus official aggregated data sources for 2017, 2018 and 2019 in the contributory regime.

https://doi.org/10.1371/journal.pone.0269079.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Comparison between the prevalence rates identified in this study versus official aggregated data sources for 2017, 2018 and 2019 in the subsidized regime.

https://doi.org/10.1371/journal.pone.0269079.t002

By using previously selected algorithms, the identified general prevalence rates in the contributory regime were 11.15(IC 95 7.87 to 14.59), 11.14 (IC95 8.78 to 13.95) and 12.37(IC 95 9.41; 14.83) per 100,000 people with the use of the sensitive algorithm for the years 2017, 2018 and 2019, respectively. Whereas prevalence rates estimated using the specific algorithm were 7.09(IC 95 5.07 to 9.40), 7.37 (IC95 5.89 to 9.35) and 8.82(IC 95 6.94; 10.55) per 100,000 people for years 2017, 2018 and 2019 respectively (Table 1).

General prevalence rates in the subsidized regime for the years 2017, 2018 and 2019 were 14.75 (IC95 1.90 to 26.34), 18.05 (IC95 2.38 to 34.70) and 15.78 (IC95 1.32 to 24.54) per 100,000 people. Using the specific algorithm, general prevalence rates were 8.77 (IC95 0.95 to 13.88), 10.48 (IC95 0.95 to 17.30) and 10.39 (IC95 0.88 to 14.21), respectively. Distribution of estimations by sex and age groups is show in Table 2.

Discussion

Using prevalence estimates is a useful strategy to design public policy and calculate attributable cost [21]. In this research two out of sixteen algorithms were selected as they had the closest prevalence rates to those rates reported by aggregated official sources. These algorithms were defined for this study as a sensitive algorithm and a specific algorithm because of their characteristics for identifying lung cancer cases, as described above. Using these two algorithms, the global prevalence rates for the contributory regime via the sensitive and specific algorithms ranged from 11.14 to 12.37 per 100,000 affiliates between 2017 and 2019 using the sensitive algorithm. In the case of the subsidized regime, general prevalence rates using the sensitive algorithm were 14.75, 18.05 and 15.78 cases per 100,000 affiliates for years 2017, 2018 and 2019, respectively.

Estimation methods used by the first two institutions were based on data obtained from samples of a limited number of patients with lung cancer attended at various providers around the country, thus introducing possible bias to the calculated rates [22, 23]. Furthermore, prevalence rates reported by Cuenta de Alto Costo were based on data provided by insurers using clinical reports as data source; although it included information from 98% of the total population affiliated with the health system, these reported data could have underestimated the real prevalence, as it depended on passive reporting [24, 25]. In the case of our research, the identification process involved the use of algorithms that take information from the usage of the reported services registered in national administrative health claims databases. This represents an advantage in comparison to the other sources, as our estimates rely on service usage rather than an active process of reporting or an estimation performed over a limited sample.

In this study, the estimations obtained by the two algorithms offer some additional advantages for policy-makers and clinicians. First, it allowed us to calculate prevalence rates differentiating between gender, regions and age groups. Second, it considered almost the total population affiliated with the contributory regime and the reported services that were utilized in the health system. Third, these estimations relied on the UPC administrative database, which represented consistent and validated data over the system, as it contains information with economical purposes. Fourth, it is possible to actively identify cases with lung cancer over the entire of population of patients attending to different providers. Finally, it is paramount to point out that these methods for identifying and estimating prevalence rates could help other LMICs to optimize their epidemiological surveillance system and to understand its impact in a context with different characteristics, compared to the developed world [26–29].

As LMIC countries enact changes to the population aging pattern, there is also a change in the burden of disease patterns from communicable to noncommunicable diseases (such as lung cancer), and efficient estimations are required to follow up on trends in public health [28, 30–32]. Authors such as Sierra et al. [33] have conducted research to address this problem and collected information from thirteen different countries in Central and South America. They found that lung cancer ranged from first to third regarding causes in incidence rates over these countries, thus showing concordance with a change in the aging patterns. In 2018, a study performed by Pilleron et al. [34] estimated that there were 679,000 new cases of cancer in Latin America and the Caribbean, and the most frequent variants were prostate, colorectal and lung cancer. While for 2020 the number of new cases estimated with lung cancer was 2,294,438 for the region according to GLOBOCAN and representing the second most frequent type of cancer [18]. The findings in this study showed a general prevalence in the contributory and the subsidized regimes ranging between 11.14 and 18.05 (using the sensitive algorithm) cases per 100,000 that positions the disease in the top ten list of prevalent cancer in the country and that corresponds to the prevalence that was identified in our study.

Another important aspect of the lung cancer prevalence rates is inequality and differences in its distribution. It has been observed that developed countries have increased prevalence rates in comparison to developing countries [35]. There is also evidence showing that at the country scale, there is a trend of higher prevalence rates within lower socioeconomic groups. This situation can be explained by differences in smoking rates or other exposure factors, such as chemical exposure, mining activities or higher exposure to air pollution that usually affects lower income population groups or states within the country [36–39]. For this study, the results showed two important patterns related to his idea. The first result showed that there is a higher prevalence rate of the disease in regions with highly polluted cities or concentrations of mining activities. The second trend found that prevalence rates that were calculated in the subsidized regime were higher than in the contributory regime. This finding follows the literature and could be explained because of the inequalities in health care access that lead to worse healthcare conditions and there fore, more case of lung cancer being detected and treated. This ideas require more research in the future.

This study had several strengths and limitations. As a strength, this research used real-world evidence from administrative claim databases that were considered to be highly reliable regarding its quality and validation procedure. Information calculated via the sensitive and specific algorithms showed results that were similar to those reported by other institutions and a range of possible prevalence rates, rather than an absolute number. In contrast, there were some concerns regarding the estimation of prevalence rates in the subsidized regime because these data were calculated based on information provided by the largest insurer in the capital city; however, this is the best information available, and it could not be easily extrapolated to the country level. There is a natural challenge in estimating real prevalence rates of lung cancer via the selected algorithms because they identify only patients with the disease and are detected by the system. However, there is a proportion of patients with the disease that are not detected by the system; hence, they were not included in our estimation. When taking into account that health coverage in Colombia reaches 99% of the total population [13], we assume that this difference was very small. Finally, it is also possible that information bias affected the study; however, quality and validation processes that were implemented by MoH helped to control this risk.

Conclusion

Selected algorithms showed similar prevalence estimations to those reported by official sources and allowed us to estimate prevalence rates in specific aging, regional and gender groups for Colombia by using national claims databases. These findings could be useful to formulate new public health strategies and to measure prevalence trends in lung cancer at a national scale by using administrative databases.

Supporting information

Acknowledgments

The authors thank Juan Camilo Forero and Juan Sebastian Castillo for their commentaries and support during the research process.