Improving the accuracy of medication adherence measures using linked prescription and dispensation data: findings from the ESOSVAL cohort of patients treated with osteoporosis drugs

Abstract

Objective: We compare estimates of proportion of days covered (PDC) based on dispensation-only data versus linked prescription and dispensation information, and we analyse their differences in a real-world cohort of patients with osteoporosis.

Methods: Prospective cohort study. We compared four alternative measures of PDC, using dispensation-only data: a) with a fixed assessment interval; b) censoring the assessment interval at the moment of the last refill; and using linked prescription and dispensation data: c) considering a minimum prescription gap of three months to interpret interruption by the physician; and d) considering any prescription gap.

Results: The mean PDC at 12 months for new users was 63.1% using dispensation-only data and a fixed interval, 86.0% using dispensation-only data and a last-refill interval, 81% using linked dispensation and prescription data and censoring any period without prescription, and 78.3% when using linked prescription and dispensation data and censoring periods of at least 3 months. For experienced users, the figures were 80.0%, 88.9%, 83% and 81%, respectively. Overall, dispensation-based measures presented issues of patient misclassification.

Conclusions: Linked prescription and dispensation data allows for more precise PDC estimates than dispensation-only data, as both primary non-adherence and early non-adherence periods, and fully non-adherent patients, are all identified and accounted for.

Keywords:

• Proportion of days covered
• medication adherence
• prescription data
• dispensation data
• methods
• real-world data

Introduction

Medication non-adherence is one of the most important barriers to getting the best benefit from pharmaceutical treatments in the real world, and to the same extent as demonstrated in clinical trials. For instance, osteoporosis medications have shown efficacy for the prevention of fragility fractures¹ but medication adherence in patients with osteoporosis has been found to be suboptimal in several studies^2–5. Poor adherence has been documented across the spectrum of chronic disease^6–16 and is associated with adverse health outcomes and higher healthcare costs⁵. Worldwide, the improvements in chronic medication initiation and adherence are at the cornerstone of policy interventions orientated towards maximizing the value of modern healthcare even if, at least to date, such interventions have usually shown mixed results¹¹.

In routine clinical practice, many factors may contribute to poor medication adherence including those related to patients, to physicians and to healthcare systems^17,18. The common belief that patients are solely responsible for taking their treatment is misleading and most often reflects a misunderstanding of how other factors affect people’s behaviour and their capacity to adhere to their treatment⁶. A better understanding of the relative contribution of patient and physician predictors of initiation, adherence and interruption of treatment can be useful for a better understanding of the complex phenomenon of non-adherence and for designing more effective interventions.

Real-world evidence on medication non-adherence is mainly based on information available in refill databases. Studies based on data which is routinely collected in the provision of care have been extremely useful for assessing adherence to and persistence with medication in patients with chronic diseases, and the impact of non-adherence and treatment interruption on clinical outcomes^7,8. However, one common feature of those studies is that they lack information about physician prescription, and adherence estimates are calculated by using dispensation data captured from pharmacy claims. When it is not possible to link prescription and dispensation data at the individual level, it is difficult to ascertain the exact moment of initiation of therapy (essential for analysing primary non-adherence) and it is also impossible to discern whether a gap in adherence may be due to patient non-adherence or to an interruption (even if temporary) of prescription as decided by a doctor. In this sense, traditional adherence estimates based on refill claims data that are not linked to prescription data, which in fact are the most prevalent in the literature, should be interpreted with caution as therapy initiation and the attribution of adherence gaps will not be reliably addressed.

In the region of Valencia in Spain, the electronic health information systems include an advanced electronic prescription manager that allows a link to be made between every patient treated in the region, prescriptions issued by doctors and the refills dispensed at the pharmacy. In this way, it is possible to overcome the aforementioned limitations of dispensation-based estimators and to calculate more refined adherence measures. To what extent adjusting adherence estimates based solely on dispensations with prescription information impacts on the adherence estimates has, to the best of our knowledge, never been explored.

In this paper we compare traditional, dispensation-based estimates with estimates using linked prescription and dispensation information, and we illustrate and quantify their differences by estimating real-world, long-term medication secondary adherence in a cohort of patients aged 50 years and over in the region of Valencia.

Methods

Design

The prospective cohort comprised the patients of the ESOSVAL cohort (fully described elsewhere¹⁹) with at least one physician prescription (to estimate secondary adherence with linked prescription and dispensation data) or one dispensation (to estimate secondary adherence with claims-only data) of an osteoporotic medication issued between June 2009 and June 2011.

Study setting

The study was conducted in the Valencia Health System (VHS), an extensive public hospital and primary healthcare centre network, which covers about 97% of the 5 million inhabitants of the Valencia region, located on the Mediterranean coast of Spain.

Population

Patients from the ESOSVAL cohort with at least one osteoporotic medication prescribed between June 2009 and June 2011 were included. The ESOSVAL cohort consists of 11,035 women and men aged 50 years and over attending 272 primary healthcare centres in the VHS for any health condition between November 2009 and September 2010. Subjects were recruited by opportunity sampling by around 600 general practitioners and primary care nurses collaborating in the ESOSVAL study following prospectively defined criteria.

We categorized patients into two groups: new users of osteoporotic treatments (when no previous dispensations or prescriptions for an osteoporotic drug were registered in the 6 months previous to the index date) and experienced users (all the rest).

Data sources

We combined data from the outpatient electronic medical record and the pharmaceutical management module of the electronic information systems of the VHS and a specific osteoporosis risk-monitoring sheet employed for the follow-up of the ESOSVAL cohort to create a database with sociodemographic and clinical characteristics and information on all physician prescriptions written and all prescriptions filled at the pharmacy for all patients studied. In the VHS, prescriptions and dispensations are linked at the individual level; treatments can be short or long term (maximum one year for chronic therapies which, for instance, would include 12 monthly prescriptions, with a window for refilling of 10 days for each prescription) and there is no monthly reimbursement limitation.

Covariates

Sociodemographic and clinical characteristics included age; sex; educational level; history of hip fracture in parents or siblings; personal history of any previous osteoporotic fracture; body mass index (BMI); falls in the last year (≥1 fall); 10 year risk of hip fracture estimated with the Fracture Risk Assessment Tool (FRAX)²⁰ and categorized into <3% and ≥3%; other secondary causes of osteoporosis; use of glucocorticoids; using the World Health Organization (WHO) osteoporosis classification criteria based on T-scores, bone mineral density results were classified as normal, osteopenia or osteoporosis²¹; sedentarism; use of calcium and vitamin D supplementation; polypharmacy (defined as having 6 or more dispensations concomitantly) and pharmaceutical copayment (categorized as no copayment for pensioners and people without resources or a copayment of 40% for the active population).

Main outcome measures

The main outcome measure was the proportion of days covered (PDC) at 12 and 24 months. PDC is generically defined as the total number of days covered with medication on hand during a specified follow-up period divided by the number of days in the patient’s follow-up period²².

We provided four alternative PDC measures. We calculated PDC with dispensation-only data in two ways, as found predominantly in the literature: a) censoring patients only in the case of death or loss to follow-up due to disenrollment, where the assessment period (almost) coincides with the follow-up period (called the “fixed interval” specification in the present study) and b) censoring the assessment periods also at the moment of the last dispensation within the follow-up period (called the “last-refill interval” in the present study). We further calculated PDC by linking prescription and dispensation data. We calculated the prescription-adjusted PDC with two different operational definitions to consider treatment interruption periods decided by the physician: c) considering a minimum prescription gap of three months to adjust PDC (meaning that, when there is a gap of three months or longer in prescription, this period is censored and is not accounted for as patient non-adherence) and d) considering any prescription gap as physician interruption (meaning that, when there is any gap in prescription, this period is censored and excluded from the calculation of the estimator). Table 1 lists the main definitions and acronyms used in this study, and Figure 1 provides illustrative examples of the calculation of PDC using the four alternative measures.

Figure 1. (a) Example of calculation of PDC using dispensation only data. PDC is calculated using either a fixed interval or a last-refill interval. (b) Example of calculation of PDC using linked dispensation and prescription data. PDC is calculated using either any gap or a three-month prescription gap.

Table 1. Definitions used in this study.

Concept	Definitions	Acronym
Primary non-adherence	Failure to have a new prescription filled (discrete event).	None
Early non-adherence	Failure to have the initial prescriptions (typically the two first prescriptions) filled (discrete event).	None
Secondary adherence	Ongoing process that measures whether or not the patient fills dispensations as prescribed during a period of follow-up and assessment.	None
Follow-up period	Total length of the period in which PDC is formally calculated, f.i., 6 months, 12 months, etc. This is the formal definition of the period in which adherence is measured, normally stated in the Title and the Methods section of the studies.	None
Assessment period	Effective period in which PDC is estimated within the follow-up period. For instance, when censoring patient time during the follow-up for any reason, the assessment period is shorter than the follow-up. If no patients are censored during the follow-up period, both follow-up and assessment periods coincide.	None
Proportion of days covered	Total number of days covered with medication during a specified assessment period divided by the number of days in patient’s assessment period.	PDC
PDC using dispensation-only data and fixed interval	PDC calculated with information on dispensations only, where the assessment period is equivalent to the follow-up period (except for reasonable censoring due to death, exclusion from insurance or other losses to follow-up unrelated to medication use).	PDC-DFI
PDC using dispensation-only data and a last-refill interval	PDC calculated with information on dispensations only, where the assessment period is censored at the time of the last dispensation within the follow-up period.	PDC-DLR
PDC using prescription and dispensation data and censoring any prescription gap	PDC calculated using linked prescription and dispensation data. Any gap in days’ supply coinciding with gaps in prescription (days not covered with prescription) is censored from the calculation of PDC.	PDC-PD
PDC using prescription and dispensation data and censoring when prescription gaps are 3 months or longer	PDC calculated using linked prescription and dispensation data. Periods equal to or longer than 3 months without prescriptions are censored and thus not included either in the numerator or in the denominator for the calculation of PDC.	PDC-PD3

f.i.: for instance.

When using dispensation-only data the index date was defined as the first dispensation, while the first prescription was used as the index date when working with linked dispensation and prescription data. This also has implications with regard to PDC calculations (see Figure 2).

Figure 2. Differences between using dispensation only data or linked dispensation and prescription data with regard to the identification of the index date and patient inclusion for PDC calculation.

Days with available medication during the follow-up period were estimated through the medication regime defined by the physician and the number of pills per package (e.g. for a regime of one pill every 12 h and packages of 30 tablets, each dispensation will entail 15 days of medication available). Stockpiling was set to 90 days. PDC was summarized categorically using the widely accepted cut-off points of PDC <20% (non-adherent), 20% ≤ PDC <80% (partially adherent) and PDC ≥ 80% (adherent or fully adherent)²³.

Osteoporosis medications

The treatments for osteoporosis included were bisphosphonates (alendronate, risedronate, ibandronate), raloxifene, bazedoxifene, strontium ranelate and parathyroid hormone/teriparatide.

Statistical analysis

We first described baseline characteristics for the whole population and then stratified by new and experienced users. Categorical variables were expressed as proportions and compared using chi-square tests.

Second, we estimated the mean PDC for new and experienced users, as well as the percentage of patients categorized as adherent, partially adherent and non-adherent at 12 and 24 months, using four alternatives for PDC estimation (see Figure 1). All analyses were conducted using STATA v13 software.

Ethics

All study subjects signed the informed consent granting researchers access to information contained in their medical record for the purposes of the study. All information was handled according to Spanish laws on confidentiality and patients’ rights. The ESOSVAL study protocol was reviewed and approved by the Committee for Ethics and Clinical Trials of the Centre for Public Health Research and the Public Health General Directorate of the Valencia Government (decision 27 March 2009, protocol modification approval 4 October 2012).

Results

Cohort characteristics

We identified 2260 patients from the ESOSVAL cohort who were prescribed an osteoporotic medication between June 2009 and June 2011; 712 (31.5%) were new users and 1548 (68.5%) were experienced users (see Table 2). For calculations based on dispensation information only, 696 new users and 1517 experienced users were considered (missing patients are those who did not fill any of their prescriptions).

Table 2. Baseline characteristics of the ESOSVAL cohort.

	New users (N = 712)		Experienced users (N = 1548)		Total (N = 2260)
	N	%	N	%
Age
50–64	380	53.4	771	49.8	1151
65–99	332	46.6	777	50.2	1109
Sex
Women	568	79. 8	1447	93.5	2015
Men	144	20.2	101	6.5	245
Education
No studies	188	29.2	496	35.2	684
Primary	287	44.6	618	43.8	905
Second/univ	169	26.2	297	21.0	366
Family history of hip fracture
No	485	82.6	1021	79.9	1506
Yes	102	17.4	256	20.1	358
Previous fracture
No	553	77.7	1258	81.3	1811
Yes	159	22.3	290	18.7	349
Body mass index
<20 kg/m^2	11	1.6	26	1.8	37
20.0–24.9 kg/m^2	190	28.2	397	26.7	587
25.0–29.9 kg/m^2	285	42.22	622	41.9	907
≥30 kg/m^2	189	28.0	439	29.6	628
Falls (≥1 in the last year)
No	486	72.4	1055	72.6	1541
Yes	185	27.6	399	27.4	584
10 year risk of hip fracturea
<3%	525	77.8	1098	74.0	1623
≥3%	150	22.2	386	26.0	536
Other osteopenic diseases^b
No	562	78.9	1241	80.2	1803
Yes	150	21.1	307	19.8	457
Glucocorticoid use^c
No	694	97.5	1515	97.9	2209
Yes	18	2.5	33	2.1	51
Other osteopenic drugs
No	407	57.16	809	52.26	1216
Yes	305	42.84	739	47.74	1044
BMD testing
No	416	58.4	979	63.2	1315
Yes	296	41.6	569	36.8	865
Bone mineral density (T-score)
Normal	30	10.3	78	14	108
−1 to −2.5	128	44.0	279	50.1	407
≤ −2.5	133	45.7	200	35.9	333
Sedentarism
No	540	79.8	1218	81.6	1758
Yes	137	20.2	274	18.4	411
Polypharmacy
≤6	269	37.8	327	21.1	596
>6	443	62.2	1221	78.9	1664
Calcium and/or vitamin D supplements
No	201	28.2	494	31.9	695
Yes	511	71.8	1054	68.1	1565
Copayment
No	533	74.9	1253	80.9	1586
Yes	179	25.1	295	19.1	474
Antiosteoporotic treatment
Bisphosphonates^d	599	84.1	1206	77.9	1805
PTH^e	10	1.4	30	1.9	40
Raloxifene	38	5.3	180	11.6	218
Ranelate	65	9.1	132	8.5	197

BMD: bone mineral density.

^aFracture Risk Assessment Tool (FRAX).

^bType I diabetes, rheumatoid arthritis, untreated long-standing hyperthyroidism, chronic malnutrition or malabsorption, chronic obstructive pulmonary disease, renal disease, prolonged immobility, organ transplantation, and chronic liver disease.

^c≥5 mg per day of prednisone or equivalent for at least 3 months in the previous year.

^dBisphosphonates (alendronate, risedronate, ibandronate).

^ePTH, parathyroid hormone (1–34 and 1–84).

Missing data: educational level (205), family history of hip fracture (396), BMI (101), falls (135), FRAX (101), sedentarism (91).

Adherence to osteoporosis medications

The mean PDC at 12 months for new users was 63.1% when using dispensation-only data and a fixed interval, 86.0% when using dispensation-only data and a last-refill interval, 81.0% using linked dispensation and prescription data censoring any period without prescriptions, and 78.3% when using linked prescription and dispensation data but censoring when prescription gaps were 3 months or longer. For experienced users, the figures were 80.0%, 88.9%, 83.0% and 81.0%, respectively. At 24 months, PDC slightly decreased using all four calculation methods for both new and experienced users (see Tables 3 and 4).

Table 3. Mean PDC and categorical PDC at 12 months using dispensation-only data and using linked prescription and dispensation data.

	New users	Experienced users
Mean PDC
Dispensation-only data	N = 696	N = 1517
PDC-DFI	63.1 (60.5–65.6)	80.0 (78.7–81.2)
PDC-DLR	86.0 (84.5–87.6)	88.9 (88.1–89.7)
Linked prescription and dispensation data	N = 712	N = 1548
PDC-PD3	78.3 (76.2–80.3)	81.0 (79.8–82.3)
PDC-PD	81.0 (78.9–83.1)	83.0 (81.7–84.2)
Categorical PDC
Dispensation-only data	N = 696	N = 1517
PDC-DFI
<20%	18.4 (15.7–21.5)	4.8 (3.9–6.1)
≥80%	46.1 (42.4–49.8)	67.0 (64.6–69.4)
PDC-DLR
<20%	1.3 (0.7–2.4)	0.07 (0.0–0.5)
≥80%	75.3 (71.9–78.4)	81.0 (79.0–82.9)
Linked prescription and dispensation data	N = 712	N = 1548
PDC-PD3
<20%	6.0 (4.5–8.0)	4.9 (3.9–6.1)
≥80%	65.0 (61.4–68.5)	70.1 (67.8–72.3)
PDC-PD
<20%	6.0 (4.5–8.0)	4.8 (3.8–6.0)
≥80%	69.7 (66.2–72.9)	73.8 (71.6–76.0)

Abbreviations. PDC, Proportion of days covered; PDC-DFI, PDC, calculated with dispensation-only data and a fixed interval of assessment; PDC-DLR, PDC, calculated with dispensation-only data and an interval of assessment censored at the moment of the last refill; PDC-PD, PDC, calculated with linked prescription and dispensation data and censoring any period of gaps in prescription from the calculation of PDC; PDC-PD3, PDC, calculated with linked prescription and dispensation data and censoring any period of 3 month gaps or longer in prescription from the calculation of PDC.

Table 4 Mean PDC and categorical PDC at 24 months using dispensation-only data and using linked prescription and dispensation data.

	New users	Experienced users
Mean PDC
Dispensation-only data	N = 696			N = 1517
PDC-DFI	56.6 (54.0–59.2)			74.1 (72.7–75.5)
PDC-DLR	81.2 (79.5–83.0)			85.7 (84.7–86.6)
Linked prescription and dispensation data	N = 712			N = 1548
PDC-PD3	77.1 (75.2–79.1)			80.3 (79.1–81.5)
PDC-PD	79.0 (77.0–81.0)			82.1 (80.8–83.3)
Categorical PDC
Dispensation-only data	N = 696			N = 1517
PDC-DFI
<20%	23.1 (20.1–26.4)			6.1 (5.0–7.5)
≥80%	35.3 (31.9–39.0)			57.0 (54.4–59.4)
PDC-DLR
<20%	1.7 (1.0–3.0)			0.7 (0.4–1.3)
≥80%	64.9 (61.3–68.4)			73.9 (71.6–76.0)
Linked prescription and dispensation data	N = 712			N = 1548
PDC-PD3
<20%	7.7 (6.0–9.9)			5.7 (4.6–7.0)
≥80%	61.7 (58.0–65.2)			67.8 (65.5–70.1)
PDC-PD
<20%	5.5 (4.0–7.4)			4.5 (3.5–5.6)
≥80%	60.3 (56.6–63.8)			68.0 (65.7–70.3)

Abbreviations. PDC, Proportion of days covered; PDC-DFI, PDC, calculated with dispensation-only data and a fixed interval of assessment; PDC-DLR, PDC, calculated with dispensation-only data and an interval of assessment censored at the moment of the last refill; PDC-PD, PDC, calculated with linked prescription and dispensation data and censoring any period of gaps in prescription from the calculation of PDC; PDC-PD3, PDC, calculated with linked prescription and dispensation data and censoring any period of 3 month gaps or longer in prescription from the calculation of PDC.

At 12 months, the percentage of non-adherent patients among new users (PDC < 20%) was 18.2% using dispensation-only data and a fixed interval, and 1.3% using dispensation-only data and a last-refill interval, and 6% when using linked prescription and dispensation data, irrespective of the gap specification. The percentage of fully adherent patients at one year for new users (PDC ≥ 80%) was 46.1% when using dispensation-only data and a fixed interval and 75.3% when the assessment period was censored at the last refill. When using linked prescription and dispensation data, PDC ≥ 80% was achieved by 65% of patients when censoring gaps in prescription equal to or longer than 3 months, and by 69.7% when censoring any gap in prescription. In experienced users, a larger proportion of patients were fully adherent irrespective of calculation methods, and similar differences in magnitude as with new users were observed among the four approaches used to estimate PDC (see Table 3). Overall, PDC-DFI underestimated patient adherence and PDC-DLR overestimated patient adherence with respect to PDC-PD. Patient misclassification using dispensation-only data versus using linked prescription and dispensation data is shown in Figure 3.

Figure 3. Patient misclassification in our cohort using dispensation data only versus using prescription and dispensation data.

Discussion

In the light of our findings, using linked prescription and dispensation data allows for a more accurate estimation of the PDC versus using dispensation data only. When prescription information is available and it is possible to link every prescription individually with every dispensation, the definition of the index date as the first prescription is more accurate than in dispensation-based studies, where the index date is usually the date of the first dispensation. In this way, the estimate is more precise as primary non-adherence and early non-adherence periods, and fully non-adherent patients (those who are prescribed in the period of assessment but do not fill any prescription) are identified and accounted for in PDC calculations (primary non-adherence in our PDC-PD cohort of new users was 6.5%). This may have important implications not only with regard to the accuracy of estimators, but for the design of interventions aimed at improving adherence and outcomes. In this way, it is possible to target high-risk patients and high-risk periods of non-adherence. Finally, an additional and important element of accuracy associated to the linkage of prescription and dispensation is the ability to censor periods without prescription for the estimation of PDC. Here, the attribution of gaps in days’ supply to patients’ non-adherence, as happens when using dispensation-only information, is imprecise, as those periods are in fact days not covered by prescription.

In our cohort of the general population of men and women aged 50 and over, we found that PDC figures based on linked prescription and dispensation data sat in the middle of those obtained from dispensation-based measures. When using dispensation-only data, differences between estimators are in turn explained by noticeable differences in the effective assessment period. When the effective period for the measurement of the PDC is censored at the time of the last refill, the effective assessment period is shortened, and thus PDC is overestimated compared to when the assessment period is a fixed interval. In fact, we could argue that this particular PDC estimator is inadequate (in the same way that some studies incorrectly censor periods in the presence of persistence gaps). On the other hand, the difference between PDC-DFI (56.6%) and PDC-PD (79%) in the case of new users shows the most noticeable numerical difference between estimators at 12 months, and reflects that, in the absence of prescription information, dispensation-based estimators using a fixed-interval period for secondary adherence assessment underestimate PDC in new users, where a phenomenon of gaps in prescription – that cannot be captured by means of dispensation information – is occurring (Appendix 1). It is commonplace in the literature related to medication adherence that naïve users tend to be less adherent to pharmacotherapy than experienced users; in the light of our results, based on higher quality information than average, this general assumption should be called into question. It is worth noting that the proportion of experienced users was not altered depending on the calculation method used, thus this factor is not affecting estimators. Finally, PDC is around 80% in most of our estimates, but a significant proportion of patients (between 25 and 65%) still remain non-adherent or only partially adherent over 24 months. This last finding is consistent with the widely reported global picture of suboptimal adherence to osteoporosis medications found in the literature, where important differences in magnitude reported among studies also arise, some showing similar figures to those of the present study^24–31.

This study has some limitations. First, our cohort was recruited by doctors that previously underwent a comprehensive, one-year long training programme in the clinical management of osteoporosis. In this sense, we may expect that our results represent high quality care occurring in the real world. This may partly explain the high values of secondary adherence to osteoporosis medications obtained in our study, although it would not affect the differences among assessment methods observed. Second, other measures of secondary adherence are used in the literature; however, PDC (or truncated MPR) is the most commonly used metric and thus more useful for comparative purposes. Third, we compare two methods of calculation of PDC based on dispensation information, while other variations can be found in the literature. For instance, some authors apply censoring when calculating PDC in the presence of persistence gaps, but we discarded these types of approaches as they are usually incorrect for adherence assessment purposes. Fourth, we did not examine the precise reasons for non-adherence, as for instance adverse effects, that may affect treatment continuation by both the physician and patient. Fifth, pharmacy claims data report only prescription filling, not actual medication use. Nevertheless, several studies have shown a high consistency between dispensation and patient consumption^32,33. Sixth, drugs dispensed at the hospital are not recorded in the database. Patients could be misclassified as non-adherent (non-persistent and low availability) during their hospital stay. Last, exposure/days’ supply misclassification may influence estimates of adherence, as shown in other contexts³⁴, although this is expected to be a minor issue in electronic dispensing systems. Despite the later limitations, retrospective cohort studies currently represent the gold standard for the estimation of real-world medication adherence.

In conclusion, we have shown that linking prescription and dispensation data allows for an accurate, refined estimation of secondary adherence versus using dispensation-only data. This offers a more complete and realistic view of real-world patient adherence, most notably in new users, where patterns of prescription interruption are quite frequent. Finally, interventions aimed at improving medication adherence may also benefit from a more accurate identification of patients at higher risk of non-adherence.

Transparency

Declaration of funding

The ESOSVAL research program is co-funded by the Instituto de Salud Carlos III (Grants PS09/02500, PI11/00238 and PI13/01721) from the Spanish Ministry of Health and the European Regional Development Fund, and by collaboration agreements established by FISABIO and the Valencia Department of Health with MSD Spain (2009–2012) and AMGEN S.A. (2010–2013), to conduct training and real-world research into musculoskeletal disorders and osteoporosis.

Author contributions

A.G., J.S., G.S. and S.P. were responsible for the study concept, design and, with I.H., for data acquisition. A.G. drafted the manuscript. I.H. prepared the database and carried out the main statistical analyses. All authors participated in the study design and interpretation of data, and contributed to the critical revision of the manuscript for important intellectual content. All authors agree to be accountable for all aspects of the manuscript.

Declaration of financial/other relationships

None of the sponsors played any role in the design of the ESOSVAL studies, the collection, analysis or interpretation of data, the writing of the manuscript or in the decision to submit it for publication. FISABIO has received research grants from various pharmaceutical companies (see funding of the ESOSVAL project). No potential conflict of interest was reported by the authors. CMRO peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgements

The authors are grateful to all doctors and nurses at the Valencia Health Agency primary healthcare centres participating in the ESOSVAL study for their collaboration, and to the Valencia Ministry of Health for its enthusiastic and continued support of the ESOSVAL research projects.

Appendix 1. Periods of effective assessment when the follow-up is defined at 24 months using different PDC calculation methods.

PDC	New users		Experienced users
	Periods of follow-up and effective assessment (months)
	Formal follow-up	Effective assessment period	Formal follow-up	Effective assessment period
Dispensation-only data
PDC-DFI	24	23.78	24	23.95
PDC-DLR	24	17.01	24	20.75
Prescription and dispensation data
PDC-PD3	24	24 (16.3)	24	24 (20.82)
PDC-PD	24	24 (15.94)	24	24 (20.38)

Abbreviations. PDC, Proportion of days covered; PDC-DFI, PDC, calculated with dispensation-only data and a fixed interval of assessment; PDC-DLR, PDC, calculated with dispensation-only data and an interval of assessment censored at the moment of the last refill; PDC-PD, PDC, calculated with linked prescription and dispensation data and censoring any period of gaps in prescription from the calculation of PDC; PDC-PD3, PDC, calculated with linked prescription and dispensation data and censoring any period of 3 month gaps or longer in prescription from the calculation of PDC.

In the case of estimators built using dispensation-only data, the effective assessment period is shorter than the formal follow-up period, due to censoring for deaths and disenrollment in the case of PDC-DFI, and to additionally censoring the time between the last dispensation and the end of the formal follow-up period in the case of PDC-PLR. The denominator of the formula for calculating the PDC is therefore shorter than the formal follow-up period. When linking prescription and dispensation data, the follow-up period remains stable (24 months), and the denominator for the calculation of the PDC is adjusted by gaps in prescription (figures in brackets).