An attempt at modeling COPD epidemiological trends in France

To optimize the allocation of resources, healthcare authorities need consistent projections estimating the future burden of major chronic diseases [1, 2]. Chronic obstructive pulmonary disease (COPD) is recognized as one of the leading non-communicable chronic diseases in terms of prevalence, mortality, morbidity, handicap and healthcare costs [3, 4]. The burden of COPD is closely related to the prevalence of the disease and to the severity distribution among COPD patients: the vast majority of COPD-related healthcare expenses (up to 70%) are related to a small fraction (< 20%) of the patients who require hospitalizations [5, 6].

Estimating the burden of COPD has proven difficult due to marked heterogeneity among epidemiological studies [7]. Methodological differences between studies relate essentially to the variability in patient sampling, and the criteria used to define COPD and to categorize its severity. Most studies were performed in specific settings such as General Practitioner (GP) practices or healthcare centers, or in limited geographical areas. Even when their results have been adjusted to the demographic characteristics of the general population, it is never possible to ensure that estimates correspond to what would be found in the “real” general population.

COPD is mostly due to tobacco smoking, which accounts for approximately 80% of cases in industrialized countries [8]. Although the role of other environmental factors is increasingly recognized [9], smoking is anticipated to remain the main risk factor of COPD for many years in industrialized countries. Consequently, trends in smoking habits in the population are considered a major determinant of the future burden of the disease in developed countries. Another important determinant of COPD prevalence is ageing, as disease prevalence increases markedly in older subjects [10].

The purpose of the present study was to use currently available epidemiological data to develop a model predicting future trends in COPD epidemiology in France, based on currently available data and to examine how variations in the main explanatory input variables associated with COPD (e.g., ageing of population, smoking habits, incidence, current prevalence, and mortality) could affect future trends.

Models developed in this study consistently predicted a modest but steady increase in COPD prevalence in France, prominently affecting women and subjects aged 75 years or more. The reference model predicted an average 0.6‰/year absolute increase in prevalence from 2005 to 2025. The variables that influenced most future trends were incidence and mortality in COPD patients. Transition rates between GOLD grades of airflow limitation had a marked influence on the prevalence of cases with severe and very severe airflow limitation.

The purpose of the present model is to allow prediction of COPD epidemiological trends over extended periods of time. Only few studies performed projections of COPD epidemiology over time. One of the first dynamic model was developed by Feenstra et al. [18] almost two decades ago and further elaborated on a few years later by Hoogendoorn et al. [13]. This model anticipated an increase in COPD prevalence for all severity stages, especially in women. In a systematic review published in 2015 by McLean et al., 6 high-quality models were identified including these two [19]. The models were designed to estimate future disease burden using trends in demographics and risk factors, Markov-type modelling and microsimulation modelling. In addition to their differences in mathematical aspects and input data, these models differ in terms of modeled output variable(s): prevalence, mortality, disability, and/or costs. These variations make it difficult to compare results between studies [7]. Therefore, sensitivity analyses represent a critical component of such modeling approaches since they allow to test the respective weight of the various input data as determinants of future disease burden. Another important aspect is external validation using actual data from longitudinal or repeated surveys. In a recent manuscript, Molinari et al. used data from hospital coding databases in France and reported increased rates of hospitalization and related-mortality [20]. Altogether, all models and surveys converge with our present study in demonstrating a high likelihood of further increase in COPD burden at a population level [3].

The initial project was planned to rely only on data collected in France, since projections were to be made for this country only, at least as a first step of the model development. In addition, it was thought that epidemiological data would be more homogeneous when restricted to a single country. However, it appeared that only a limited number of epidemiological studies were available, some of which were ancient. Thus, some of the input data were obtained using articles published in the same time frame in industrialized countries, which were available in a limited number of countries, resulting in some heterogeneity. In addition, methods of case recruitments were quite different: in one study, the studied population was recruited in health care prevention centers and results were adjusted to match the characteristics of the general population [11]. In that study, an 8.4% prevalence of COPD was found, which is consistent with the results of other epidemiological studies in developed countries. However, the proportion of patients with severe airflow obstruction was rather low, in contrast to what had been observed elsewhere [7]. The authors hypothesized that this finding could be biased by the source of the sample, since patients visiting prevention centers for a general health check-up are supposed not to be managed for some severe illness. In a European study (the European Community Respiratory Health Survey), part of the studied population was recruited in France [21]. However, only two towns participated, which makes it difficult to extrapolate to the whole French population, even after appropriate adjustments; in addition, the age structure of the population recruited in that study was very different (age range, 20–44 years). In the Confronting COPD study [22] and its follow-up study [23], to which France participated [24, 25], the diagnosis of COPD did not rely on spirometry but only on subjects-reported medical history and symptoms. Considering the possible bias in the only spirometric study available in the French general population [11] and the noticeable heterogeneity in available results regarding COPD severity distribution, it was decided to perform sensitivity analyses using data from various studies performed in developed countries at roughly the same time the French reference data were collected. These analyses suggested that, provided the initial prevalence estimates are robust, incidence and mortality rates are the main determinants of prevalence projections. Therefore, it appears important to include these variables in standard population surveillance programs, to which they do not belong at present in many countries. Our results also emphasize that regularly gathering reliable prevalence estimates at the national level is crucial to refine projections of future disease burden. Another question of interest is the contribution of occupational, domestic and atmospheric pollutants to the occurrence, severity and natural history of COPD. At present there are very few longitudinal data on how these exposures evolve by region over time and how they modulate trends in COPD epidemiology.

A surprising finding was the predicted decrease in the number patients in GOLD 1, which was counter intuitive. Because this decrease was accompanied by a comparable increase in the number of patients in GOLD 2, we speculate that this finding may be related to imbalance between the number of new GOLD 1 patients vs. a higher number of patients who transit from GOLD grade 1 to grade 2. However, we cannot exclude the presence of unidentified bias in the model.

Models such as the one developed here can produce data to inform estimations of future disease-related costs and thereby to guide future resource allocation. As previously shown, the structure of a model developed using mostly data from one country could be applied to another region provided that adequate input data is available [26]. Another application of these modeling approaches is the assessment of the cost and cost-effectiveness of long-term care [27, 28]. However, as recently emphasized in a systematic review, a major difficulty with health economics simulation models in COPD is to fully account for the disease’s heterogeneity and to include the weight of comorbid conditions [2]. Most general population epidemiological studies to not provide sufficient levels of details on the clinical characteristics of identified patients. These details can be found in observational studies or clinical trials, but corresponding populations are unlikely to represent the COPD population at large.

In conclusion, epidemiological models based on current estimates can project future COPD burden. Their reliability is heavily dependent on the quality of input data. Therefore, regular surveillance of COPD epidemiology is required to obtain robust updated projections that can be used to guide healthcare policies.