Letter to the editor

Re: Ulcickas Yood M, Oliveria SA, Cziraky M, et al. Adherence to treatment with second-line therapies, dasatinib and nilotinib, in patients with chronic myeloid leukemia. Curr Med Res Opin 2012, 28(2): 213-19.

Dear Editor,

In a recent paper published in CMRO, Ulcickas Yood et al. presented an analysis based on the HealthCore Integrated Research Database comparing real-world treatment adherence in a population of patients diagnosed with chronic myeloid leukemia (CML) receiving dasatinib or nilotinib as a second-line therapy. The topic of this study is of substantial interest to readers, particularly those working in the area of CML treatment. However, we feel the analysis, though proclaiming some significant results, is based upon a method lacking theoretical foundation. The authors defined and estimated adherence in an unconventional manner without first establishing the validity and reliability of the method. Based on our assessment, we believe the approach applied in the study could lead to erroneous conclusions. Moreover, there is lack of clarity and transparency in the description of the method that has been used to obtain these results, especially given the unconventional approach.

Firstly, in the Patients and methods section under the subsection Data analysis, the authors indicated that they used the medication possession ratio (MPR) as a measure of adherence and mentioned that “Patients were considered to have poor adherence if their MPR was <85%”. The MPR was calculated as the number of days of supply of the current prescription divided by the total days between the current and next prescription. Cox proportional-hazard models were used to quantify the rates of poor adherence. This non-conventional method suggests that patients would be adherent until a certain time-point (failure) after which the patient is considered as non-adherent. However, no specification was provided for the definition of the ‘failure’ date. With the MPR being measured as the number of days supply of the current prescription divided by the total days between the current and next prescription, it is unclear if the date of the failure was the date of the ‘next’ prescription fill or the end date of the supply for the current prescription. The definition of failure is critical in the interpretation of the Cox proportional-hazard model; hence, some clarification of this step is of utmost necessity.

Based on a personal communication with Dr Ulcickas Yood, Dr Chen has confirmed that the authors measured the MPR repeatedly at each prescription fill, based on the gap between two consecutive prescription fills. According to their non-adherence definition, if there is a gap of greater or equal to 6 days between the end date of the period of supply and the next prescription fill date for a typical 30–day supply prescription, the patient would be considered as non-adherent. This method is conceptually incorrect as adherence is not a point-estimate measure, but a measure that should be taken over a given period of time^, In particular, we feel that this method is inappropriate for a claims database analysis, where no accurate information is available on when the medication was actually taken. Let us consider the following examples representing typical cases that could be frequently observed in a clinical setting:

1. Assume we have two patients. The first patient filled his/her second prescription 6 days late (for any reason), but this patient had been taking his/her medication on time since then for another 2 years. The second patient filled the second prescription within less than 6 days (e.g., 5 days after the end of the first prescription) and after the end of the second prescription, this patient discontinued the therapy. The time-to-non-adherence method used by the authors would consider the first patient to have worse compliance compared to the second patient.

2. Another example is the case of a patient who leaves for vacation. When returning from vacation, a patient may refill the prescription 6 days later, and even if the patient is perfectly compliant to medication after this 6–day gap, the patient will be considered as non-adherent. Paradoxically, in the event that the patient had foreseen the difficulty of filling the prescription on time while on vacation and had decided to refill the second prescription 6 days before the end of the first prescription, as a result, the third prescription may have been refilled 6 days after the end of the period of supply; in this case, using the authors’ method, because of extra storage the patient has carefully planned to avoid non-adherence, the patient would be classified as non-adherent starting from the second prescription.

The above examples reflect the two main unconventional aspects of the method used in the study. First, a 6–day gap is much shorter than the typical 30- to 90–day gap applied in most of the therapy adherence literature. Second, adherence was measured over a very short period of time (between a first prescription and the next prescription), rather than over an extended period of time. The combined effect of these two limitations leads to a measure that is extremely sensitive to common noises, as presented in the above examples, and that is not a reliable measure of the overall adherence. We question the value of the development and the use of such measure, particularly when used in combination with a survival analysis approach.

There are various medical and non-medical reasons why a gap of at least 6 days would be observed between two prescription fills and, measuring the MPR solely based on the gap between two prescription fills (the method used by the authors) would result in a large group of patients being classified as non-adherent while these patients were actually adherent to their medication. Fundamentally, the authors require patients to refill all prescriptions perfectly on time, with very small room for variation (not earlier or later). This is an unrealistic expectation and unprecedented in the compliance study literature and leads to conclusions mainly based on random events rather than based on systematic differences.

We would like to thank the editors for their time and consideration. We look forward to further clarification to address the potential issues raised in this letter.

Sincerely,

Eric Q. Wu

Analysis Group, Inc. Boston, MA, USA.

Lei Chen

Novartis Pharmaceuticals, Corp. East Hanover, NJ, USA

Authors’ response to the Letter to the Editor

Marianne Ulcickas Yood

Susan Oliveria

Mark Cziraky

Dear Editors,

Thank you for forwarding comments on our paper, “Adherence to treatment with second-line therapies, dasatinib and nilotinib, in patients with chronic myeloid leukemia1”. We appreciate the opportunity to respond.

In their letter, Wu and Chen describe our methods as “non-conventional”; our methods include medication possession ratio (MPR) and Cox modeling. However, as stated and referenced in our paper, “…although no single measurement of adherence…can be considered a gold standard approach, MPR is considered a suitable proxy for measuring adherence when using automated data,”2,3 and MPR has been used as an adherence measure in studies of other oral cancer drug4,5. As well, Cox modeling has been widely used since the early 1970s6 and is considered a fundamental approach when evaluating outcomes (in this case non-adherence) over time. We agree that the definition of failure is important and clarify that we measured MPR at each prescription fill date, and defined failure as the point in time at which a patient fulfilled the definition of non-adherence (MPR <85%).

Our approach in measuring MPR is not “conceptually incorrect” as stated by Wu and Chen but rather should be considered as a thorough evaluation of small gaps in treatment refills. We consider evaluating small gaps in treatment refills to be a conservative approach, with the understanding that achieving and maintaining clinical response is closely related to adherence8–11. The assertion proposed by Wu and Chen, that a hypothetical 6–day gap could drive the results for patients, is not supported by the data presented in our paper. Examination of Figure 1 in our paper1 clearly demonstrates that we evaluated continuous prescriptions over a long time period, and the curves begin to separate (i.e., indicating differences in adherence) at 100 days for nilotinib vs. dasatinib ≤100 mg/day, and continue to separate over time (with follow-up extending over 1 year).

In the Discussion section of our paper, we outlined the potential limitations of our study. However, one of the most important factors for the readers to understand is that this is a comparative assessment of two drugs used to treat the same condition. As a result, any theoretical critiques regarding the gap used or the method used to define adherence are not germane unless refill or vacation patterns differ between patients exposed to nilotinib and dasatinib, as postulated in the letter submitted by Wu and Chen.

We appreciate the opportunity to further clarify the findings in our paper. These data identify a difference in adherence between dasatinib and nilotinib, demonstrating that in real-world clinical practice there are important and meaningful differences that warrant consideration when making treatment decisions.

Sincerely,

Marianne Ulcickas Yood

Susan Oliveria

Mark Cziraky

Muhammad Hamdan

Notes

* Ulcickas Yood M, Oliveria SA, Cziraky M, et al. Adherence to treatment with second-line therapies, dasatinib and nilotinib, in patients with chronic myeloid leukemia. Curr Med Res Opin 2012, 28(2):213–19

† Peterson AM, Nau DP, Cramer JA, Benner J, Gwadry-Sridhar F, Nichol M. A checklist for medication compliance and persistence studies using retrospective databases. Value Health 2007;10(1):3–12

‡ Cramer JA, Roy A, Burrell A, et al. Medication compliance and persistence: terminology and definitions. Value Health 2008;11(1):44–7