Ability to pay for medication: a clustering analysis of 1404 patients with the Patient Financial Eligibility Tool

Overall, anonymized economic data from 1404 patients were extracted for the present analysis, including 947 from Thailand, 347 from UAE and 110 from Mexico. Mean normalized (0–100) PFET summary component scores for income, assets and SoL were 10 ± 19, 16 ± 14, 19 ± 11 for Thailand, 13 ± 20, 3 ± 7, 22 ± 13 for UAE and 31 ± 21, 10 ± 12, 25 ± 13 for Mexico, respectively.

Figure 1 shows the results from the clustering analysis using the SOM methodology in Thailand (A), UAE (B) and Mexico (C). For each series of SOMs by country, patients identified as globally similar in terms of PFET income, assets and SoL component scores are grouped in ‘districts’ represented as numbered polygons placed at fixed positions on the maps. The more those specific patient subgroups resemble each other in terms of PFET values, the closer they are on the map and conversely, subgroups with fewer commonalities are further apart. Colors ranging from blue to red indicate average district value levels, from the lowest to highest. For illustrative purpose, a selection of average district values is shown for representative districts in each SOM.

After visual analysis of all the SOMs across countries, seven subtypes of patients were apparent, according to the following possible combinations between the three PFET metrics: globally wealthy patients (cluster 1), characterized by increased values in all three PFET metrics and represented in all three countries in the left part of the SOMs; patients with elevated SoL and income (cluster 2), characterized by increased values in those two metrics but not in assets, and found in UAE (upper area) and Mexico (left area); patients with elevated SoL and assets (cluster 3), found in Thailand (lower left area) and Mexico (lower area); patients with isolated elevated assets (cluster 4), only found in Thailand (upper area); patients with isolated elevated income (cluster 5), only found in Mexico (upper area); patients with isolated elevated SoL (cluster 6), found in Thailand (lower area) and UAE (lower area); and globally poor patients with low values in all three metrics (cluster 7), found in all three countries in the right part of the SOMs. Using this classification, a global map indicating the proposed clusters labels is shown for each country in the left area, whereas other maps on the right show the distribution of PFET summary component scores with the clusters boundaries superimposed using solid black lines. Detailed characteristics of the clusters with mean (standard deviation) of quantitative illustrative features are shown in Table 1, along with p-values from global comparison tests. Distribution of clusters across countries varied from 11 to 17% (unweighted average 14%) for Cluster 1 (globally wealthy), from 42 to 56% (unweighted average 48%) for cluster 7 (globally poor), and from 32 to 44% (unweighted average 38%) for the remaining in-between clusters 2–6. Systematically significant differences between clusters were found regarding the three PFET metrics, with higher standardized values (/100) found in Cluster 1 (i.e., from 48 to 55 for income, from 16 to 31 for assets and from 27 to 40 for SoL) and the lowest values in cluster 7 (i.e., from 2 to 14 for income, from 1 to 6 for assets and from 11 to 17 for SoL).

Three-dimensional biplot visualization based on PC analysis is shown in Supplementary Video 1 for the three countries, further illustrating the differentiated characteristics of the patients across the clusters. Globally wealthy patients (cluster 1) projected on the same directions as the three PFET variables, whereas globally poor (cluster 7) projected on opposite ones. Intermediate profiles (clusters 2–6) projected according to the combinations previously described.

The results presented here demonstrate that the PFET is a valuable tool for identifying and characterizing the economic status of specific patient subgroups in different countries. Importantly, these findings also underscore the country-specific differences in the correlation between the specific economic indicators of income, assets and SoL and patients’ overall economic status. For example, wealth in Mexico is more directly tied to income than it is to SoL, whereas assets capture some clusters of patients (shown in orange) (Figure 1C). This demonstrates that the combination of income and assets is an effective way to assess wealth status in Mexico. However, SoL seems to be the primary driver of wealth in the UAE, whereas SoL and assets play a larger role in wealth in Thailand (Figure 1A & B). These findings demonstrate the potential pitfalls of using any one economic indicator as a single metric across all countries, and supports the use of a multi-factorial approach, such as the PFET in determining patients’ financial eligibility to pay for patented drugs and, potentially, to be able to manage other types of financial expenditures.

Another key finding is that the specific types of clusters that can be identified using data collected by the PFET also vary among the three countries included in the analysis. For example, both the UAE and Thailand have a cluster of patients defined by a higher SoL (Figure 1A & B & Table 1), whereas this cluster is not found in Mexico (Figure 1C & Table 1). Similarly, Thailand has a cluster of patients defined by higher assets, whereas Mexico and the UAE do not. This again highlights the importance of both utilizing multiple indicators to determine economic status and tailoring the indicators used to reflect the economic factors within individual countries.

In fact, economic factors varied among the three countries included in this analysis, and the results provide several examples of the interplay between them and the ability to effectively assess patients’ wealth status. For example, the patient population assessed in Mexico resides in structured, urban economics settings in which the informal economy has a limited impact. In contrast, the informal economy is a very important factor in Thailand, which makes income an unreliable indicator for determining wealth status in this country. Instead, SoL and, to some extent, assets are the most reliable indicators for Thai patients. The UAE has yet other factors that must be considered in assessing patient wealth status and ability to pay for medication. This is due to the fact that most patients in the UAE who need financial assistance for purchasing medication are expatriates whose assets are primarily located outside of the country. Additionally, some patient clusters have high SoL without high income, which may be due to workers receiving weekly wages in cash. Again, these diverse economic environments further underscore the utility of examining multiple metrics of wealth assessment rather than relying on a single indicator that may not be relevant in all countries or communities.

It should be noted that the PFET utilizes indicators of wealth as its metrics, whereas the financial assessment tools typically used by charitable organizations evaluate indicators of poverty. This may account in part for the variation in income, assets and SoL among the countries included in this analysis compared with what might be expected based on poverty indicators within these countries.

The ‘Numero’ cluster analysis algorithm used in this study combines the self-organizing map algorithm, permutation analysis for statistical evidence and a final expert-driven subgrouping step and has previously been shown to enable the identification of subgroups with related features despite the lack of an obvious clustering structure [9]. The current study provides additional evidence that cluster analysis can be a powerful tool for evaluating a variety of patient characteristics and gaining insights into the clinical and economic factors that impact patients’ access to care, disease management strategies and overall health outcomes.

Finally, our study has limitations worth discussing, including the lack of comparison to other asset-based wealth indices such as the International Wealth Index [11] or the Comparative Wealth Index [12], as data collection was not initially designed to capture the same items as those required for calculation of these indices and we consequently lacked targeted or sufficiently detailed information in our database to draw comparisons. Second, we did not evaluate the correlation between PFET results and macroeconomic development measurements at the national level (e.g., Gross National Income per capita or poverty indicators from World Bank such as Poverty Headcount Ratios [11]) to provide additional validation of the predictive ability of PFET. Such analysis would have been limited by the number of countries analyzed in the present work (n = 3) and could have yielded potentially misleading results given the selection of our sample, which was not constituted based on a representative sampling methodology but based on the real-life enrollment of patients in drug access programs.

The continuing rise in the cost of prescription medications, coupled with growing and aging populations, places an increasing burden on governments around the world to meet the healthcare costs of their citizens. A growing proportion of patients in low- and middle-income countries face the challenge of purchasing expensive and life-saving medications while trying to balance other essential expenses and not descend down the economic ladder. Strategies utilizing modest price reduction and uniform free-of-charge treatment have been pursued, but their overall health impact has limited because they often do not allow patients to receive a full course of medication therapy that is needed for optimal outcomes. The PFET is a versatile tool that can be adapted to each country’s economic context to assess patients’ ability to contribute to their medication costs and determine the subsidy they need to complete their treatment course. By allowing subsidy funds to be used more effectively, the PFET may play an important role in ensuring continuity of treatment and improving access to prescription medicines for more patients.

Ongoing discovery and innovation in clinical medicine is a powerful driver for more specialized and increasingly individualized healthcare. Continued advance of this trend is likely to culminate in more widespread adoption of personalized therapies tailored to each patient’s genetics in the next 5–10 years. Such individualized approaches will require more individualized care, treatment and follow up and likely involve more care overall. Ensuring that patients in low- and middle-income countries have access to and benefit from newer, more personalized treatments will require effective use of these patients’ financial resources and the financial resources of patient support programs. The ability to accurately assess patients’ financial ability to contribute financially to the cost of their prescription medications will thus become even more critical in developing sustainable access models. Consequently, PFET will be an increasingly valuable tool in this evolving health system context.

E Audureau has contributed to the conception of the work, data analysis and interpretation, and to drafting the manuscript. MH Besson contributed to the conception of the work, data acquisition and interpretation of the results, and to revising critically the manuscript. A Nofal, R Jayasundera, J Saba and J Ladner contributed to the conception of the work, interpretation of the results and to revising critically the manuscript. All authors gave their final approval of the manuscript and agreed to be accountable for all aspects of the work.

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Writing assistance has been used for the creation of the manuscript and funded by Axios International.

An informed and signed consent has been obtained from all participants for data collection and treatment. All analyses were performed based on anonymized information.

This work is licensed under the Attribution-NonCommercial-NoDerivatives 4.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-nd/4.0/