Short Communication
For reprint orders, please contact: reprints@futuremedicine.com
Ability to pay for medication: a clustering
analysis of 1404 patients with the Patient
Financial Eligibility Tool
Etienne Audureau*,1, Marie H ´
el `
ene Besson2, Anas Nofal3, Roshel Jayasundera4,
Joseph Saba2&Jo
¨
el Ladner5
1Public Health Department, H ˆ
opital Henri Mondor, Assistance Publique Hˆ
opitaux de Paris, 51 av du Mar´
echal de Lattre de Tassigny,
94000, Cr´
eteil, France; CEpiA EA7376, University Paris Est Cr ´
eteil
2Axios International, 20 Rue Cambon, 75001 Paris, France
3Axios International, DSC Tower, 7th Floor, Unit 709, Dubai Studio City, PO Box 500767, Dubai, UAE
4Axios International, Regal Tower, 11th Floor, Suite 1101, Business Bay, PO Box 24005, Dubai, UAE
5Rouen University Hospital, Epidemiology & Health Promotion Department, H ˆ
opital Charles Nicolle 1, rue de Germont, 76031
Cedex Rouen, France
*Author for correspondence: Tel.: +33 1498 13664; Fax: +33 1498 13697; etienne.audureau@aphp.fr
Aim: The study was conducted to understand how key determinants of the Patient Financial Eligibility
Tool (PFET), a previously validated tool for assessing patients’ ability to contribute to their medication
costs, vary across countries. Materials & methods: A clustering analysis was conducted on economic data
from 1404 patients from Thailand (n =947), the UAE (n =347) and Mexico (n =110). Results: The analysis
identied seven patient clusters, including globally wealthy or poor patients (14%/48%) and those with
only selectively increased PFET economic indicators (38%), and revealed country-specic differences in the
correlation between PFET metrics and patients’ overall economic status. Conclusion: 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.
First draft submitted: 22 May 2019; Accepted for publication: 21 June 2019; Published online:
12 July 2019
Keywords: ability to pay •low-to-middle income countries •unsupervised clustering analysis •wealth index
A key challenge to increasing access to patented drugs for patients in low- and middle-income countries is identifying
and implementing sustainable payment models. Donation programs established by pharmaceutical companies may
address one-time or short-term needs for vaccines and medications but may not be optimized for creating access
to medications that require chronic use. In the absence of free medications, most patients in these countries pay
for them out of pocket, which may also not be sustainable when the medications must be purchased for long-term
treatment of chronic illnesses. An alternative and potentially more sustainable approach to making patented drugs
available in low- and middle-income countries is a hybrid model in which patients with some financial resources
pay a portion of their medication costs, with the remaining costs covered to ensure that patients take full course of
treatment and maximize medical benefits.
Such a hybrid model could provide multiple benefits that would expand access to life-saving patented medications.
First, it could allow a broader segment of patients across the economic spectrum to afford these medications. Second,
a larger number of patients contributing more to the cost of their medications, to the extent they can afford, could
allow resources provided through free medication and other subsidized programs to be more effectively deployed
in helping those patients who lack the economic resources to purchase the full course of treatment that is necessary
for optimal therapy.
Key to enabling hybrid models is determining the amount that patients can contribute to the cost of their
medications without significant economic distress or deleterious financial burden, with additional costs covered
through financial support programs. Although income has been used as an indicator of financial wealth and
expenditures have been used as a proxy for income [1], these factors may not accurately reflect a patient’s real
J. Comp. Eff. Res. (Epub ahead of print) ISSN 2042-630510.2217/cer-2019-0063 C
2019 Axios International
Short Communication Audureau, Besson, Nofal, Jayasundera, Saba & Ladner
economic status, especially in countries with informal and/or cash economies. Income, in particular, has been
found to be difficult to capture accurately, as it may tend to fluctuate substantially depending on seasonality or
migration, it does not account for informal earnings, and it is a sensible information which may prove difficult to
collect from individuals [1,2]. Wealth metrics may provide a more accurate assessment of a patient’s ability to pay
for medication, but these data may rely on the collection of numerous questions which adds to survey length and
are thus not commonly collected [2,3]. Additionally, the accuracy of any individual wealth index may vary based on
the variables included and the weights and values assigned to questions and responses for specific indicators [3,4].
Asset-based methods to assess economic status are also limited due to their inability to account for the short-term or
temporary economic changes, including the changes associated with addressing a significant healthcare expense [4,5].
In order to address the limitations associated with the independent use of income, expenditures or wealth
to determine economic status, we previously developed and validated a novel multilateral index that combines
income, asset and standard of living (SoL) data [6]. This index, the Patient Financial Eligibility Tool (PFET), is
adapted on a country-by-country basis to address the cultural, behavioral and financial standards within individual
countries. Although patients’ individual wealth level as measured by the tool was previously validated against their
actual ability to pay for expensive treatments [6], it is yet unknown to what extent the instrument may operate
differently across countries, for example, how income information may be the most important driver in some
countries while SoL/assets may more strongly influence the final result in other countries characterized by greater
informal economies. Similarly, the question arises as to whether there are distinct patterns of combinations of the
three metrics that could be used to identify specific patients’ subgroups or profiles. Such information is crucial to
documenting the adaptability of the PFET in varying settings and, ultimately, establishing its utility for routine
wealth level assessment within the hybrid model previously described. To that end, use of unsupervised machine
learning techniques (clustering) may provide important insights into existing patterns of wealth metrics by reducing
complexity and facilitating the identification and interpretation of the combinations at play among income, assets
and SoL.
In our ongoing effort to adapt the PFET to variation in local country context and to maintain its efficacy as a
tool for assessing patients’ ability to contribute to the cost of patented medications, we conducted a cluster analysis
using detailed data from patients enrolled in drug access programs using the PFET in Mexico, Thailand and the
UAE. The goal of this analysis was to understand how the key determinants of PFET vary across countries. Here,
we report the results of this cluster analysis, which was designed to identify and characterize subgroups of patients
who have common economic status as determined by the different component metrics of the PFET.
Material & methods
Study population & data collection
Since its initial development, the PFET has been progressively deployed in a number of low- to middle-income
countries (N = 20 as of 2019) for a wide variety of drugs prescribed in the field of chronic diseases or cancer (N = 57
as of 2019). A dedicated database platform has been developed in parallel with the use of the PFET to enable routine
collection of patient economic data as well as to enable internal testing to adjust the tool and optimize its ability to
quantify patients’ ability to pay. For the present analysis, data were extracted from three countries (Thailand, UAE
and Mexico), which were chosen based on the availability of data within the database and because the size of the
informal economy varies across them [7]. Consequently, these three countries provide an excellent sample population
in which to evaluate interactions among the component metrics of the PFET in distinct economic environments.
Anonymized data were extracted for patients enrolled in 12 different drug treatment programs between May 2013
and February 2017.
Information used for the present analysis included the following variables: normalized (0–100) PFET summary
components for income, assets and SoL based on the linear weighted combination of all individual items relating
to these topics, and detailed quantitative items expressed in purchasing power parity dollars to allow cross-country
comparisons, including patients’ household and economic unit income, monthly household expenditures, financial
assets and, when applicable, vehicle price, value/rental of primary dwelling, business turnover and education fees.
Qualitative information collected within the PFET regarding household equipment, method of cooking, water
source or dwelling floor material was not included because of the variability of the subcategories used across
countries. Clustering analysis was conducted based on the normalized three PFET metrics (income, SoL, assets),
whereas detailed quantitative information were used as illustrative features to describe the main characteristics of
the clusters identified.
10.2217/cer-2019-0063 J. Comp. Eff. Res. (Epub ahead of print) future science group
Clusters of ability to pay for medication Short Communication
Statistical analysis
Clustering analysis relied on self-organizing maps (SOMs), a nonparametric neural network clustering technique
that allows the visual identification of homogenous groups (clusters) in 2D maps. Patients sharing similar char-
acteristics in terms of income, assets and SoL are displayed in close proximity on the SOMs, whereas patients
with opposite features are mapped in areas distant from one another [8]. A circular implementation applying the
‘Numero’ statistical framework [9] was used to build the SOMs and to define clusters boundaries, by constructing
the SOMs with statistical verification of the robustness of the contrasts observed through permutation tests and
determining suitable groupings of patients based on the visual identification of key data patterns. All analyses were
stratified by country.
For illustrative purpose, a Gabriel’s biplot [10] was created to project the patients from the three countries along
the principal components (PCs) axes from a PC analysis based on the three PFET metrics. Cluster solutions were
then mapped on the biplot by coloring patients according to their cluster membership. Descriptive quantitative
results are presented as means (±standard deviation). Comparisons between groups identified from clustering
analysis were conducted using one-way analysis of variance (ANOVA) or Kruskall–Wallis tests for continuous data,
as appropriate. A p-value <0.05 was considered significant.
Analyses were performed using Stata 15.1 (StataCorp, TX, USA) for descriptive analyses and between-groups
comparisons, and R 3.4.2 statistical software (Vienna, Austria; Numero and pca3d packages) for clustering analyses
and visualizations.
Results
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)
future science group 10.2217/cer-2019-0063
Short Communication Audureau, Besson, Nofal, Jayasundera, Saba & Ladner
Globally wealthy
Clusters Income Assets Standard of living
Clusters
Thailand
UAE
Mexico
Income Assets Standard of living
Clusters Income Assets Standard of living
Globally poor
Higher standard of living/assets
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
20
19
18 17 16
15
14
1
2
3
4
5
6
7
813
12
9
21
34
35
42
41
65
66
36 37 38 39
40
43 44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
10 11
33
32
31
30
27 26
25
24
23
22
28
29
60
61
62
63
64
Higher standard of living
Higher assets
Higher income
Globally wealthy
Globally wealthy
Globally poor
Globally poor
Higher standard of living/income
Higher standard of living/assets
Higher standard of living/income
Higher standard of living
+33
+44
+32
+27
+7.9
+2.71
+0.96
+4.39
+26
+24
+14.9
+9.7
+20
+18 +5.9
+15
+35
+32
+12.9
+43
+60
+72
+49
+12.2 +4.45
+19 +14.1
+24
+8.5
+2.84 +22
+50
+31
+41
+33 +21
+11.8
+14.4
+16
+4.84
+2.55
+1.27
+8.7
+20
+26
+28
+3.80
+28
+2.51
+1.47
+6.3 +0.099
+26
+11.5
+10.9
+26
+0.060
+0.082
+0.139
+38
+71
+0.093 +6.7
+19
+0.149
+2.22
+0.58
+2.10
+0.090
+0.095 +35
+40
+27
+37
+31 +15
+9.9
+1.69
+1.77
+0.085
+1.46
+60
+0.0398
+27
+5.7
44
+16
Figure 1. Results from the clustering analysis by self-organized maps according to country. Self-organized map placed patients with similar economic status within small
groupings (‘districts’) throughout the map. Each individual map shows the average values per district for each economic indicator, in other words, income, assets and standard of
living, with blue colors indicating the lowest values and red colors the highest. Based on visual identication of key patterns in the self-organized maps, close districts were
combined to provide suitable clusters of patients whose boundaries are indicated delimited by solid black lines.
10.2217/cer-2019-0063 J. Comp. Eff. Res. (Epub ahead of print) future science group
Clusters of ability to pay for medication Short Communication
Table 1. Main economic characteristics of patients according to clusters and country.
Characteristics Cluster 1
Globally wealthy;
mean (±SD)
Cluster 2
Elevated Standard
of living/income;
mean (±SD)
Cluster 3
Elevated standard
of living/assets;
mean (±SD)
Cluster 4
Elevated assets;
mean (±SD)
Cluster 5
Elevated income;
mean (±SD)
Cluster 6
Elevated standard
of living; mean
(±SD)
Cluster 7
Globally poor;
mean (±SD)
p-value
Thailand N=131 (14%) N=98 (10%) N =170 (18%) N=146 (15%) N=402 (42%)
Income (/100) 48 (±28) 8 (±6) 3 (±5) 5 (±6) 2 (±4) ⬍0.001*
Assets (/100) 31 (±15) 25 (±6) 31 (±10) 6 (±6) 6 (±5) ⬍0.001*
Standard of living (/100) 27 (±11) 34 (±6) 19 (±7) 26 (±7) 11 (±6) ⬍0.001*
Median patient’s household and
economic unit income ($PPP; IQR)
8711 (7105; 12,544) 4041 (2718; 5508) 2885 (1443; 3607) 2525 (1324; 4208) 2164 (1082; 3511) 0.015*
Vehic le ($PPP) 28,378 (±24,929) 60,713 (±208,260) 42,140 (±110,178) 49,390 (±198,054) 22,664 (±98,504) 0.040*
Ownership of primary dwelling ($PPP) 287,370 (±276,403) 447,628
(±1,129,678)
361,425
(±1,308,345)
435,252
(±3,130,555)
146,931 (±590,921) 0.142
Business description ($PPP) 13,391 (±68,289) 2411 (±12,147) 3684 (±19,254) 2613 (±22,375) 992 (±11,054) 0.001*
Financial assets ($PPP) 53,927 (±240,953) 366,151
(±3,112,416)
37,025 (±98,021) 17,586 (±83,621) 13,336 (±79,926) 0.036*
Rental of the primary dwelling ($PPP) 101 (±496) 67 (±281) 71 (±429) 168 (±670) 189 (±1546) 0.687
Education fees ($PPP) 460 (±929) 668 (±2558) 233 (±594) 544 (±2519) 389 (±2494) 0.509
Monthly household expenditures ($PPP) 1155 (±1117) 1958 (±5261) 1781 (±6649) 1759 (±8392) 1389 (±4739) 0.737
UAE N=39 (11%) N =42 (12%) N =70 (20%) N =196 (56%)
Income (/100) 52 (±31) 32 (±7) 7(±6) 3 (±5) ⬍0.001*
Assets (/100) 16 (±13) 2 (±3) 1(±2) 1 (±3) ⬍0.001*
Standard of living (/100) 37 (±10) 33 (±10) 31 (±5) 13 (±8) ⬍0.001*
Median patient’s household and
economic unit income ($PPP; IQR)
9724 (7001; 14,780) 6807 (5834; 7973) 3112 (1556; 4278) 1964 (895; 2937) ⬍0.001*
Vehic le ($PPP) 8471 (±9089) 2812 (±2754) 2797 (±4952) 831 (±1945) ⬍0.001*
Ownership of primary dwelling ($PPP) 44,998 (±171,763) 0 (±0) 0(±0) 595 (±7076) ⬍0.001*
Business description ($PPP) 4787 (±29,895) 0 (±0) 0(±0) 0 (±0) 0.047*
Financial assets ($PPP) 15,999 (±24,746) 2249 (±2857) 1298 (±2352) 1455 (±8906) ⬍0.001*
Rental of the primary dwelling ($PPP) 2928 (±5842) 2009 (±1317) 1370 (±760) 515 (±536) ⬍0.001*
Education fees ($PPP) 795 (±1169) 693 (±1288) 615 (±1142) 105 (±214) ⬍0.001*
Monthly household expenditures ($PPP) 607 (±356) 557 (±292) 494 (±353) 202 (±179) ⬍0.001*
*p ⬍0.05.
$PPP: Purchasing power parity dollar; IQR: Interquartile range; SD: Standard deviation.
future science group 10.2217/cer-2019-0063
Short Communication Audureau, Besson, Nofal, Jayasundera, Saba & Ladner
Table 1. Main economic characteristics of patients according to clusters and country (cont.).
Characteristics Cluster 1
Globally wealthy;
mean (±SD)
Cluster 2
Elevated Standard
of living/income;
mean (±SD)
Cluster 3
Elevated standard
of living/assets;
mean (±SD)
Cluster 4
Elevated assets;
mean (±SD)
Cluster 5
Elevated income;
mean (±SD)
Cluster 6
Elevated standard
of living; mean
(±SD)
Cluster 7
Globally poor;
mean (±SD)
p-value
Mexico N=19 (17%) N =12 (11%) N =10 (9%) N =19 (17%) N=50 (45%)
Income (/100) 55 (±17) 47 (±16) 24 (±13) 44 (±10) 14 (±8) ⬍0.001*
Assets (/100) 23 (±8) 7 (±5) 28 (±22) 6 (±5) 3 (±3) ⬍0.001*
Standard of living (/100) 40 (±15) 42 (±6) 30 (±8) 22 (±6) 17 (±7) ⬍0.001*
Median patient’s household and
economic unit income ($PPP; IQR)
958 (452; 1773) 931 (188; 1773) 616 (0; 887) 709 (213; 1419) 288 (67; 709) 0.013*
Vehic le ($PPP) 2333 (±4703) 296 (±1024) 1552 (±4753) 1027 (±3201) 319 (±1590) 0.125
Ownership of primary dwelling ($PPP) 127,400 (±92,533) 48,397 (±35,640) 118,370 (±92,241) 67,667 (±72,167) 29,531 (±26,280) ⬍0.001*
Business description ($PPP) 1027 (±4475) 0 (±0) 8867 (±19,154) 0 (±0) 62 (±380) 0.001*
Financial assets ($PPP) 5500 (±6660) 924 (±1325) 6588 (±15,083) 1177 (±2099) 490 (±1353) 0.001*
Rental of the primary dwelling ($PPP) 83 (±290) 111 (±384) 0 (±0) 80 (±194) 61 (±146) 0.796
Education fees ($PPP) 133 (±246) 219 (±506) 8 (±25) 118 (±227) 37 (±95) 0.068
Monthly household expenditures ($PPP) 532 (±212) 612 (±212) 350 (±120) 472 (±225) 319 (±160) ⬍0.001*
*p ⬍0.05.
$PPP: Purchasing power parity dollar; IQR: Interquartile range; SD: Standard deviation.
10.2217/cer-2019-0063 J. Comp. Eff. Res. (Epub ahead of print) future science group
Clusters of ability to pay for medication Short Communication
projected on opposite ones. Intermediate profiles (clusters 2–6) projected according to the combinations previously
described.
Discussion
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, permu-
tation 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.
future science group 10.2217/cer-2019-0063
Short Communication Audureau, Besson, Nofal, Jayasundera, Saba & Ladner
Conclusion
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.
Future perspective
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.
Executive summary
•Patient Financial Eligibility Tool (PFET) as a tool for assessing patients’ ability to pay for prescription medications.
•Due to its use of multiple component metrics, the PFET is a sensitive tool for assessing patients’ economic status
and ability to pay for prescription medications.
•Cluster analysis of PFET data gathered from patients in three countries identied distinct subpopulations that
share or lack common characteristics.
•The ndings also underscore the country-specic differences in the correlation between the specic economic
indicators of income, assets and standard of living and patients’ overall economic status.
•Effectively determining patients’ ability to contribute to their prescription medication costs may allow subsidy
funds to be used more effectively, thus providing more patients with access to a full course of medical therapy
that can improve their healthcare outcomes.
Authors’ contributions
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 nal approval of the manuscript and agreed to be accountable for
all aspects of the work.
Financial & competing interests disclosure
The authors have no relevant afliations or nancial involvement with any organization or entity with a nancial interest in or nan-
cial conict 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.
Ethical conduct of research
An informed and signed consent has been obtained from all participants for data collection and treatment. All analyses were
performed based on anonymized information.
10.2217/cer-2019-0063 J. Comp. Eff. Res. (Epub ahead of print) future science group
Clusters of ability to pay for medication Short Communication
Open access
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/
References
Papers of special note have been highlighted as: •of interest
1. Rutstein O, Johnson K. The DHS Health Index. DHS Comparative Reports 6 (2004).
https://dhsprogram.com/pubs/pdf/CR6/CR6.pdf
2. Chakraborty NM, Fry K, Behl R, Longfield K. Simplified asset indices to measure wealth and equity in health programs: a reliability and
validity analysis using survey data from 16 countries. Global Health Sci. Pract. 4(1), 141–154 (2016).
3. Davies JB. An overview of personal wealth. In: Personal Wealth From A Global Perspective, Davies BJ (Ed.). Oxford University Press, NY,
USA, 1–23 (2008).
4. Filmer D, Pritchett LH. Estimating wealth effects without expenditure data – or tears: an application to educational enrollments in states
of India. Demography 38(1), 115–132 (2001).
•Provides key background information on wealth index development.
5. Vyas S, Kumaranayake L. Constructing socio-economic status indices: how to use principal components analysis. Health Policy
Plan 21(6), 459–468 (2006).
6. Saba J, Audureau E, Bize M, Koloshuk B, Ladner J. Development and validation of a multilateral index to determine economic status in
developing countries: the Patient Financial Eligibility Tool (PFET). Popul Health Manag. 16(2), 82–89 (2013).
•Provides information on the Patient Financial Eligibility Tool initial development and validation.
7. Schneider F, Buehn A, Montenegro CE. Shadow economies all over the world: new estimates for 162 countries from 1999 to
2007. Policy Research Working Paper. The World Bank (2010).
http://documents.worldbank.org/curated/en/311991468037132740/pdf/WPS5356.pdf
8. Kohonen T, Self-organizing maps (3rd Ed.) In: Springer Series in Information Sciences. Huang ST, Kohonen T, Schroeder RM.
(Eds.). Springer, NY, USA (2001).
9. Gao S, Mutter S, Casey A, Makinen VP. Numero: a statistical framework to define multivariable subgroups in complex population-based
datasets. Int. J. Epidemiol. doi:10.1093/ije/dyy113 (2018) (Epub ahead of print).
•Provides key information on the Self-Organized Maps methodology and its implementation within the Numero statistical
framework.
10. Gabriel KR. The biplot graphic display of matrices with application to principal component analysis. Biometrika 58, 453–467 (1971).
11. Smits J, Steendijk R. The International Wealth Index (IWI). Soc. Indic. Res. 122(1), 65–85 (2015).
12. Rutstein O, Staveteig S. Making the demographic and Health Surveys Wealth Index comparable. DHS Methodological Reports
9(2014). www.dhsprogram.com/pubs/pdf/MR9/MR9.pdf
future science group 10.2217/cer-2019-0063
