Use of pharmacy records to measure treatment adherence : a critical review of the literature

The current frame of reference on adherence to pharmacotherapy includes a set of behaviors experienced by the user, with observation of the detailed and continuous history of the use of each dose of the medication. Indicators based on pharmacy records have been used to measure adherence. The current review aimed to identify and describe indicators based on pharmacy records and to discuss their adequacy and limitations for measuring adherence. An exploratory literature review was conducted in three databases using the terms “adherence”, “pharmacy records/administrative data”, and “measure” to compose the descriptors for the selection of 81 articles and the elaboration of a chart with the denomination, sources, methods for calculation, description, and interpretation of the operational and referential meaning of 14 indicators. Given the most recent taxonomy for adherence proposed in the literature, we concluded that the indicators can be useful for identifying patients with medication-seeking behavior-related problems and analysis of persistence. The distance between supply-related events and difficulties in treatment followup can influence an analysis based exclusively on the use of these indicators. Medication Adherence; Indicators of Health Services; Pharmaceutical Services Correspondence E. C. Lima-Dellamora Rua Gregório de Castro Morais 860, Rio de Janeiro, RJ 21931-350, Brasil. lima.dellamora@gmail.com 1 Faculdade de Farmácia, Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brasil. 2 Escola Nacional de Saúde Pública Sergio Arouca, Fundação Oswaldo Cruz, Rio de Janeiro, Brasil. REVISÃO REVIEW doi: 10.1590/0102-311X00136216 Cad. Saúde Pública 2017; 33(3):e00136216 This article is published in Open Access under the Creative Commons Attribution license, which allows use, distribution, and reproduction in any medium, without restrictions, as long as the original work is correctly cited. Lima-Dellamora EC et al. 2 Cad. Saúde Pública 2017; 33(3):e00136216 Adherence to drug therapy can be included in the scope of the overall concept of adherence adopted by the World Health Organization (WHO) 1 (p. 3), which defines it as “the extent to which a person’s behavior – taking medication, following a diet, and/or executing lifestyle changes, corresponds with agreed recommendations from a health care provider”. The theoretical framework of the WHO is broad, since it considers multiple determinants of adherence, from those farthest from the patient/user (includes the health system and services) to the closest, such as issues related to the medication. In order to justify measuring adherence, some criteria need to be met, such as a prescription based on a proven therapeutic need, correct dose indication, dosing regimen, treatment time, and availability of an adequate amount of the medication 2. The literature presents and classifies the methods for measuring adherence as direct and indirect. Measures based on direct criteria are considered the most reliable, since they are based on objective verification of intake (directly observed therapy) or on plasma titration of the drug 2,3. Meanwhile, indirect measures – resulting from interaction between the user and instruments for measuring adherence (e.g., interviews and self-completed questionnaires) – are susceptible to recall bias or exercise of the patient/user’s will 2,4. In order to create conditions of objectivity, the information resulting from the interview or questionnaire is backed by scales, which transform the data into measurable items 4,5. The scales include elements on medication-seeking behavior, medication-taking behavior, beliefs, and barriers to adherence and their determinants. Not all scales measure adherence under the same clinical conditions or in the same contexts 4. Other indirect measures like physical or electronic counts of available medications for use attempt to generate data on frequency of intake. Although more objective than interviews or self-report, it is not possible to guarantee that the counted medications are actually taken. The current taxonomy on adherence to medication includes a set of behaviors by the user 6 and emphasizes the need to seek answers in the “dosing history”, translated as the detailed and continuous trajectory in use of the medication, at each dose 7. In this sense, adherence could be defined as a stage in the multifactor regressions between the drug’s prescription and the therapeutic effect 8,9. In the final analysis, measuring adherence is intended to predict a clinical outcome: the medication’s effectiveness 10. In studies on adherence, secondary databases can be useful by offering the possibility for quick access to a set of individualized information from a large number of users. However, the clinical context is sometimes indispensable to compare and determine the information’s validity 11. Considering the limited availability of clinical and prescription records, many studies are limited to national or local information systems with records on the supply and dispensing of medications 12,13,14,15. These records have been used to construct indicators for the analysis of adherence and its determinants 12,16,17,18,19,20,21,22. The current study thus aimed to identify and characterize indicators based on pharmacy records and discuss their adequacy and limitations for measuring adherence, based on a literature review. The decision was made to address methodological aspects related to the measurement of adherence, with a focus on analyzing the applicability of these indicators and theoretical, conceptual, and practical assumptions that justify and provide the basis for this measurement. Methodological approach We conducted an exploratory literature review in PubMed, SciELO, and Scopus, using the terms “adherence”, “pharmacy records/administrative data” and “measure” to compose the descriptors and boolean equations [OR, AND], allowing the selection of articles published in English, Spanish, and Portuguese, with no limitation on year of publication. Next, the following exclusion criteria were applied in two stages (reading the abstracts and full texts), as shown in the flowchart (Figure 1): (i) duplicate articles, (ii) full text unavailable, or (iii) articles that did not address adherence and/or did not use administrative data or pharmacy records. Information from the literature supported the elaboration of a descriptive chart of the indicators, including sources and application, method of calculation, description, and interpretation of the indicator’s operational meaning and frame of reference. USE OF PHARMACY RECORDS TO MEASURE TREATMENT ADHERENCE 3 Cad. Saúde Pública 2017; 33(3):e00136216 Figure 1 Flowchart for article selection process. The chart revealed the implications for use of the indicators and precautions in interpretation of the results, assuming that it is not possible to study adherence without being certain that the patient received the medication in the prescribed quantity and at the right time 2,23. We attempted to define aspects related to treatment coverage and possession versus gaps in the medication as ways of exploring the possible proximal or distal relations between availability of the medication, use, and adherence. Finally, we drew on the theoretical and conceptual framework of adherence drawing on “dosing histories”, based on which a new taxonomy was developed, which strengthens the concept of adherence as therapeutic result 2,4,6,7,24. Three reasons were weighed on the adequacy and limits of the indicators which provide the basis for measuring adherence in practical terms: (i) evaluation of a drug intervention’s therapeutic effects; (ii) determination of the influence of adherence on specific clinical conditions; and (iii) identification of patients that need orientation or support to use their medications better 4. These reasons and the above-mentioned theoretical concepts provided the basis for analyzing the indicators and orienting the discussion. Lima-Dellamora EC et al. 4 Cad. Saúde Pública 2017; 33(3):e00136216 Results Indicators based on the dispensing process have been developed over the years in research and health service settings, based on practical criteria – given the greater availability of administrative or supply records than prescription data or consumption per se 8,25,26,27,28. These indicators (i) focus on the dynamics of individual supply (dispensing of the medication), (ii) translate critical information on the amounts supplied (number of units/doses), and (iii) allow verifying the supply timeline. The method calculates the periods, generally in days, in which one assumes the patient’s possession of the medication following its dispensing or the possible gaps resulting from irregular receipt of the drug. Fourteen studies, among the 81 retrieved, addressed the relationship between pharmacy records and adherence, but did not present descriptive measures (indicators) of this relationship 18,29,30,31,32,33, 34,35,36,37,38,39,40,41. Some 40 studies pointed to measures (indicators) based on supply records to measure adherence 42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79, 80,81,82,83,84,85 and were mostly published before Vrijens et al. 6 proposed a new taxonomy. The literature refers to the principal indicators used to measure adherence in various types of chronic diseases as medication possession ratio (MPR) and proportion of days covered (PDC). The studies diverge as to their usefulness in the strict evaluation of adherence 40,42. However, there is use of these indicators to measure treatment adherence in mental disorders 43,44, depression 45, schizophrenia 46,47,48,49,50, diabetes 50,51,52,53,54,55, hypertension 51,56,57,58,59,60,61,62, coronary diseases 63,64,65,66,67, osteoporosis 68,69, rheumatoid arthritis 70,71, lupus 70, chronic obstructive pulmonary disease 72,73,74, asthma 75,76,77,78, HIV/AID

Adherence to drug therapy can be included in the scope of the overall concept of adherence adopted by the World Health Organization (WHO) 1 (p. 3), which defines it as "the extent to which a person's behavior -taking medication, following a diet, and/or executing lifestyle changes, corresponds with agreed recommendations from a health care provider".
The theoretical framework of the WHO is broad, since it considers multiple determinants of adherence, from those farthest from the patient/user (includes the health system and services) to the closest, such as issues related to the medication. In order to justify measuring adherence, some criteria need to be met, such as a prescription based on a proven therapeutic need, correct dose indication, dosing regimen, treatment time, and availability of an adequate amount of the medication 2 .
The literature presents and classifies the methods for measuring adherence as direct and indirect. Measures based on direct criteria are considered the most reliable, since they are based on objective verification of intake (directly observed therapy) or on plasma titration of the drug 2,3 .
Meanwhile, indirect measures -resulting from interaction between the user and instruments for measuring adherence (e.g., interviews and self-completed questionnaires) -are susceptible to recall bias or exercise of the patient/user's will 2,4 . In order to create conditions of objectivity, the information resulting from the interview or questionnaire is backed by scales, which transform the data into measurable items 4,5 . The scales include elements on medication-seeking behavior, medication-taking behavior, beliefs, and barriers to adherence and their determinants. Not all scales measure adherence under the same clinical conditions or in the same contexts 4 .
Other indirect measures like physical or electronic counts of available medications for use attempt to generate data on frequency of intake. Although more objective than interviews or self-report, it is not possible to guarantee that the counted medications are actually taken.
The current taxonomy on adherence to medication includes a set of behaviors by the user 6 and emphasizes the need to seek answers in the "dosing history", translated as the detailed and continuous trajectory in use of the medication, at each dose 7 . In this sense, adherence could be defined as a stage in the multifactor regressions between the drug's prescription and the therapeutic effect 8,9 . In the final analysis, measuring adherence is intended to predict a clinical outcome: the medication's effectiveness 10 .
In studies on adherence, secondary databases can be useful by offering the possibility for quick access to a set of individualized information from a large number of users. However, the clinical context is sometimes indispensable to compare and determine the information's validity 11 . Considering the limited availability of clinical and prescription records, many studies are limited to national or local information systems with records on the supply and dispensing of medications 12,13,14,15 . These records have been used to construct indicators for the analysis of adherence and its determinants 12,16,17,18,19,20,21,22 .
The current study thus aimed to identify and characterize indicators based on pharmacy records and discuss their adequacy and limitations for measuring adherence, based on a literature review. The decision was made to address methodological aspects related to the measurement of adherence, with a focus on analyzing the applicability of these indicators and theoretical, conceptual, and practical assumptions that justify and provide the basis for this measurement.

Methodological approach
We conducted an exploratory literature review in PubMed, SciELO, and Scopus, using the terms "adherence", "pharmacy records/administrative data" and "measure" to compose the descriptors and boolean equations [OR, AND], allowing the selection of articles published in English, Spanish, and Portuguese, with no limitation on year of publication. Next, the following exclusion criteria were applied in two stages (reading the abstracts and full texts), as shown in the flowchart ( Figure 1): (i) duplicate articles, (ii) full text unavailable, or (iii) articles that did not address adherence and/or did not use administrative data or pharmacy records.
Information from the literature supported the elaboration of a descriptive chart of the indicators, including sources and application, method of calculation, description, and interpretation of the indicator's operational meaning and frame of reference.  Flowchart for article selection process.
The chart revealed the implications for use of the indicators and precautions in interpretation of the results, assuming that it is not possible to study adherence without being certain that the patient received the medication in the prescribed quantity and at the right time 2,23 . We attempted to define aspects related to treatment coverage and possession versus gaps in the medication as ways of exploring the possible proximal or distal relations between availability of the medication, use, and adherence.
Finally, we drew on the theoretical and conceptual framework of adherence drawing on "dosing histories", based on which a new taxonomy was developed, which strengthens the concept of adherence as therapeutic result 2,4,6,7,24 . Three reasons were weighed on the adequacy and limits of the indicators which provide the basis for measuring adherence in practical terms: (i) evaluation of a drug intervention's therapeutic effects; (ii) determination of the influence of adherence on specific clinical conditions; and (iii) identification of patients that need orientation or support to use their medications better 4 . These reasons and the above-mentioned theoretical concepts provided the basis for analyzing the indicators and orienting the discussion.

Results
Indicators based on the dispensing process have been developed over the years in research and health service settings, based on practical criteria -given the greater availability of administrative or supply records than prescription data or consumption per se 8,25,26,27,28 .
These indicators (i) focus on the dynamics of individual supply (dispensing of the medication), (ii) translate critical information on the amounts supplied (number of units/doses), and (iii) allow verifying the supply timeline. The method calculates the periods, generally in days, in which one assumes the patient's possession of the medication following its dispensing or the possible gaps resulting from irregular receipt of the drug.
In general, the parameters for analysis of the amount (or doses) of the medication and the periodicity of expected dispensing were the medical prescription or therapeutic protocol 86,87 . Based on the prescription, the ideal therapeutic behavior projected over time generated an estimate of the medications to be consumed during the period. Based on this ideal scenario, one verifies the translation of the idea of "coverage" or "percentage of coverage" 8,28 .
The concept of coverage has been operationalized by calculating the days' supply, which represents the amount (in days of treatment) that the total medication covers, in one or more episodes of dispensing, over the period (in days) theoretically covered by the amount dispensed. Often, any observed gap is expressed as the difference between the total observation period and the number of days' supply in possession.
In addition to the possession or lack of the medication, the development of the indicators was based on the adoption of two other aspects, in distinct combinations, that gave each measure's specificity: (a) whether the indicator's result is treated as a dichotomous or continuous variable and (b) whether the verification occurs in one or multiple dispensing intervals 8 .
While not sharing the same calculation method, the indicators represented different expressions of the same measurement logic 78 . In the numerator, this logic could be stated as "days' supply of medication obtained" or "number of days in which the patient was in possession of the medication as prescribed", as expressions of the treatment supplied/obtained. In the denominator, "observation period" (in days) or "period between the first and latest dispensing" (in days) emerged as expressions of the ideal coverage time. Since they contained the same unit in the numerator and in the denominator, the results were adimensional.
Continuous single-interval of medication availability (CSA) and continuous multiple-interval of medication availability (CMA), as well as continuous single-interval of medication gap (CSG) and continuous multiple-interval of medication gap (CMG), are examples of measures that sought to reflect the balance between the analysis of single or multiple episodes of dispensing. The expression of the result for a single dispensing (in the case of CSA, CSG) appeared useful for indicating the fraction of coverage in acute diseases. However, for the analysis of adherence in the continuous treatment of chronic diseases, the proposal has been to use the mean of the measurements of single intervals  Table 1 Indicators based on pharmacy supply records. Measures the sum of time intervals in which the patient had a given medication available in a series of dispensing intervals.

Measure/Sources
Aims to express coverage in multiple serial periods and is generally expressed as a fraction.
Values greater than 1 mean "oversupply". Measures the single interval or multiple (total) time intervals in which the patient has the medication available, like CSA and CMA, respectively.
Usually expressed as a fraction.
For cases of polypharmacy, it is suggested to consider the mean of the values for the numerator before dividing by the denominator. Values greater than 1 mean "oversupply", but some studies suggest that MPR should be truncated at 1.
The indicator is sometimes expressed as a ratio whose antecedent is the fractional result of the calculation and the consequent is always 1.  For a single dispensing, it is the time interval in which the patient did not have the medication available (dispensing gap). In periods with more than one dispensing, one can use the mean (mean gap in a given interval).
Since the observation period is set arbitrarily, it is not based on the real history of use. If the observation period is the period between the first and the last, the measure has fewer limitations.
There are limitations in the cases of "oversupply". In cases with excess, the measure can only be used for multiple cases (patient population) in which the negative values (referring to excess, in calculating the gap) can be redefined as 0, as presented in the literature.  In series of dispensing intervals, the sum of the time intervals in which the patient did not have the medication available (total gap).
Periods of "oversupply" involve negative values in the indicator, usually redefined as 0.
It is a simplified measure that does not have the power to estimate the number of days on which the patient was without the medication ("real" gap). Not based on the dosing history. It is like a 'mean'. as the expression of continuous availability. However, this would involve a loss of information on coverage gaps in sequential periods, disguising coverage gaps in the past with medication surpluses in the future 8,26 . Another relevant aspect relates to the definition of the observation interval in the denominator. The studies treated the terms "interval" and "days evaluated" as synonymous. In Table 1, the denominator in the indicators CMA, CMG, MPR, PDC, medication refill adherence (MRA), and continuous multiple-interval measure of over-supply (CMOS) is described as the "observation period" and denotes an arbitrarily defined period, dates for the start and finish of the data collection, in which the target variables are examined. Meanwhile, the definition of the denominator as the "period between the first and last dispensing", as in the indicators refill compliance rate (RCR), compliance rate (CR), and days between fills adherence rate (DBR) appears to have avoided the arbitrary definition of the observation interval. In these indicators, the limit of the observation period was defined as the date of a dispensing episode.
Finally, in the numerator and/or denominator, the indicator may or may not include the amount furnished in the last dispensing in the days' supply. Inclusion in the numerator would mean the possibility of higher results than unity (or greater than 100%). Values greater than one or one hundred percent may or may not mean oversupply.
As shown in Table 1, the proposal for some indicators resulted from the attempt to refine the underlying logic, based on the empirical data, to deal with difficulties in comparing the expected and the observed values. This is especially true of the indicators that inserted a critique or adjustment to refine the results obtained. These adjustment strategies (or mathematical devices) occurred in the numerator of RCR, CR, and DBR, in the denominator of medication possession ratio modified (MPRm), or in the expression of the final result (truncated MPR).

Discussion
The analysis of the specific variations between the methods for each indicator's calculation allowed understanding the operational limitations involved in the extraction and standardization of the data and the identification of necessary precautions when interpreting the results.
The search strategy allowed identifying studies that incorporated indicators based on supply records as measures of adherence, for the construction of Table 1. Other studies retrieved in the search defined adherent patients based on administrative supply records, but either did not describe in detail the indicator employed, or used complementary methods to validate the measure of adherence 18,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102 . The latter feature the methodological care in using more than one strategy to obtain information on adherence, although not detailing specific measures for its assessment (the focus of this review).
The supply records in the studies on use of medications generate simple, low-cost, non-invasive measures that dispense for the use of questionnaires/interviews 11,86,87 . In computerized systems, information is concentrated in databases or software programs 87 , facilitating data access and organization. When the medication is supplied by a single agency or entity with a computerized system, the patient's history can be viewed rapidly 81,87 .
The inference between the dynamics of pharmacy supply of medications expressed in indicators based on supply records and adherence is conditioned on three assumptions. The first is that the database on dispensing includes all the sources for obtaining the prescribed medications. The second is that lack of supply of medications in the amounts and on the proper dates implies the impossibility of use as prescribed 86 . It is not possible to study adherence without the certainty that the patient received the medication in the adequate amount 2,23 . The third is that if the medications were dispensed in the expected period, they were consumed as prescribed 3,81,86 . Another quite radical assumption is necessary -that everything supplied was taken at the prescribed amounts, at the prescribed hours, and for the prescribed time -that is, when estimating adherence based exclusively on supply data, one assumes perfect adherence to all the other unmeasured aspects.
In addition, the use of indicators based on supply records as measures of adherence requires careful consideration of the reasons behind the measure in clinical practice 4 .
The first issue focuses on the discussion of which relations are possible between the medication's supply and its therapeutic effects. The literature claims the indicators' pertinence as the measure of adherence based on empirical demonstrations of their predictive validity for clinical outcome and their correlation with alternative methods for measuring adherence 29,39,40,43,44,56,70,103,104,105,106,107,108,109 . However, despite their usefulness as predictors of target clinical outcomes, the assertion does not imply the validity of their use as measures of adherence. The point here is not to question such indicators based on the inherent limitations of the sources employed and the method for calculation, but to discuss what can be reasonably inferred from the results.
Adherence has traditionally been estimated either as a dichotomous indicator or as a percentage, in which the patient's medication-taking behavior is recorded by the health professional, measured on a scale from 0 to 100 108,110 . To transform the interval measures in dichotomous, cutoffs points are set for the respective disease according to the literature, which requires caution. The variable's dichotomization prevents distinguishing between types of non-adherence: consistent or sporadic; patients at different stages in the medication-taking process, etc.
Vrijens et al. 6 debated the phenomenon of adherence after conducting a review on the topic and producing a consensus with 80 experts from 13 European countries, based on the study of the drugdosing histories. The authors proposed to define adherence as a process consisting of three components: initiation, implementation, and discontinuation. The components correspond respectively to use of the first prescribed dose, the dynamic of maintaining the treatment regimen (length, frequency, and hours of dose administrations), and the definitive interruption of use, for various reasons. Permeating these phases is an attribute known as persistence, described as the period of time between initiation and the last dose 6,42,108,111 .
Persistence is a measure associated with the time of uninterrupted maintenance of the treatment regime and can be measured through the episodes of the patients' return to pick up their medications, Cad. Saúde Pública 2017; 33(3):e00136216 over time. It is an important outcome for evaluating treatment adherence and has been displayed visually as Kaplan-Meier curves, like survival 6,7,66 , or inferred as a measure resulting from the use of the PDC indicator 42,69,100,112 . However, although persistence is related to adherence -and is sometimes confused with it in the literature 65,99,112 -, persistence is not synonymous with adherence, nor is it an unequivocal predictor of it according to dosing history. Treatment persistence is possible without adherence to the prescribed recommendations. Therefore, the most satisfactory measure of adherence is that which includes persistence and all the stages defined in the new paradigm of adherence 7 .
Quantification of implementation requires comparison of two time series: the prescribed treatment regimen, which can be compared to the expected parameter, and the patient's treatment history, corresponding to the observed parameter 6 . The result of measuring implementation can be expressed by a single summary measure (normally a percentage of adherence at the end of a defined period)which is the most common in practice -or by a longitudinal sequence of measurements 110 . Supporting the former is the practicality of classifying discrete outcomes (adherent versus non-adherent) on a predetermined scale of relative distances between the expected and the observed. However, the same result, or percentage of implementation, can mean completely different medication-taking behaviors, that is, treatment regimen histories, or differentiated adherence, with direct implications for the medication's effectiveness 7 . At the limit, differences exist in effectiveness, even among patients that follow their treatment regimen correctly (and which are thus adherent to it), expressing the so-called "unreliable link" between use of the medication and the therapeutic effect 7 .
Thus, the first aspects to be considered in the sequence of explanatory events that lead to the body's response to the treatment regimen are not those related to possession of medication or continuity or volume of the supply, but those involving ingestion of the dose, followed by pharmacokinetic aspects, which directly explain the therapeutic effects, based on the drug's absorption and metabolism 7,113,114 . In this sense, establishing a correlation, even if indirect, between the indicator based on records and supply and the therapeutic effect means ignoring a series of intermediate elements in the dosing history that were not investigated.
To understand the factors that influence adherence under specific clinical conditions, it is necessary to characterize not only the time sequence of events involving the treatment regimen, but their determinants. Since adherence involves a series of phases and attributes -initiation, implementation, discontinuation, persistence 6 -the factors associated with some of these attributes are not necessarily associated with all of them 24 . It is necessary to investigate specific factors related to the different components of adherence. The indicators reviewed here do not allow investigating how the treatment was initiated or discontinued, or the characteristics of its implementation. Meanwhile, persistence allows the incorporation of time limits on the implementation of adherence and can thus be measured by the indicators. However, there are limitations on the measure of persistence that are inherent to the use of secondary databases 11,44,47,51,64,69,115 . In a systematic review, Kardas et al. 24 found 771 different variables for expressing factors associated with adherence to prolonged therapies. Most were related to implementation, while 47 were related to persistence. This discussion should conclude by highlighting the usefulness of indicators based on supply records. Osterberg & Blaschke 2 contended that supply data on medications can be an estimate of the extent to which the patient persists with the treatment. However, as discussed, one should not assume that possession of the medication or persistence in retrieving it, per se, mean adherence. At any rate, lack of the medication certainly constitutes a factor for non-adherence. Therefore, the relevance of these indicators is clear in the identification (screening) of patients that need orientation or support to improve their use of medications for chronic diseases 10 , especially when associated with other methods 116 .
In Brazil, most studies involving the use of indicators based on supply records focus on antiretroviral therapy 12,18,117 , available in the System for Logistic Control of Medications (Siclom), which combines prescription with supply (but without addressing all the stages of dispensing). Recording in this database is mandatory for nearly all of the pharmacies accredited for antiretroviral dispensing. Some studies that used Siclom combined pharmacy records with different sources of information (self-report, patient diary, pharmacy records, and laboratory data) to measure the percentage of nonadherent patients, focusing on supply gaps. Cardoso & Galera 17  the time interval recorded between dispensing episodes for analysis of the use of medications and treatment dropout, respectively. A more recent study, on biological therapy in rheumatic diseases, specified the use of the PDC indicator to identify patients without adherence to treatment 21 . Other studies from Brazil and elsewhere have focused on persistence as the outcome measure for use of indicators based on pharmacy records, which suggests evolution in adherence concepts as the result of dosing history 56,95,100,112 .

Final remarks
This review presented a map of indicators reported in the literature, used to analyze adherence to pharmacotherapy. The strategy to build a descriptive picture of indicators based on pharmacy supply records allowed identifying the precautions and limitations in studies on the use of medications that can contribute to the design and comparability of future studies in the field.
As discussed, the indicators, stemming from different evaluations of the supply dynamics in medications, based on pharmacy records, resulted from the refinement of a single logic: that the time of exposure to treatment was calculated from the amount of medications supplied for an assumed time period. This improvement and the resulting variations reflect the operational difficulties in obtaining and use of the measures.
By expressing the fraction (adimensional) equivalent to a time segment of the treatment, the indicators' results refer more adequately to the "persistence" component, in the taxonomy adopted for adherence. They are thus adequate as outcome measures in analyses of factors associated with this component of adherence. These indicators are also clearly useful for identifying patients with problems related to medication-seeking behavior, indicating lack of adherence.
It is true that traditionally, studies on the use of medications employ supply data to estimate consumption. However, the gap between supply-related events and difficulties in treatment follow-up that impact consumption disallow relating the results of the indicators with clinical aspects, essential in the discussion of adherence.