Data source
The Japan Medical Data Center (JMDC), a for-profit company (www.jmdc.co.jp), collected health insurance claims of employees and their family members from several health insurance societies in Japan [10]. Using these data, the JMDC has created an anonymous and individually traceable database using their patented technology of anonymous linkage of individual claims. This database contains the following information collected by the JMDC for each claim: diagnosis information (unique Japanese disease name, International Classification of Diseases and Related Health Problems, 10th revision (ICD-10) code, and diagnosis date in year/month); prescribed drug information (drug name, dose, duration, prescription date in year/month/day); enrollee’s characteristics based on their health insurance (sex, birth year, dates of insurance enrollment, and disenrollment); and dates of individual-specific data collection periods. The data set used in this study was extracted from a total of over 3 million enrollees between 2005 and 2014 (comprising 1,640,000 employees and their 1,450,000 family members), including individuals who did not use any healthcare services.
The Research Ethics Committee of the Graduate School of Medicine, The University of Tokyo in Japan approved the study protocol (Protocol No. 10724-(1)).
Study design
The study design was a cohort study of patients with overactive bladder who had not used the treatment drugs previously. The study period consisted of a 6-month run-in period (May 1, 2010 to November 18, 2010) and a subsequent 3-year analysis period (November 19, 2010 to November 18, 2013).
The inclusion criteria were as follows: patients with overactive bladder aged 20 years or older, who had been diagnosed with overactive bladder before May 1, 2010 and whose medical information was continuously collected by the JMDC throughout the study period. We excluded patients who had been prescribed a treatment drug during the 6-month run-in period (May 1, 2010 to November 18, 2010).
The definition of overactive bladder was that a patient had the unique Japanese disease name of overactive bladder, kakatsudoboko, with an ICD-10 code of N32.8 in their medical claims records. Treatment drugs were defined as anti-cholinergic drugs (solifenacin, imidafenacin, tolterodine, propiverine, oxybutynin, or flavoxate) or a beta-3 adrenergic receptor stimulator (mirabegron) based on the clinical guideline of the Japanese Continence Society at the study period [11]. Solifenacin and mirabegron are products of Astellas. Fesoterodine and darifenacin were not available for use in Japan during the study period.
Advertising exposure
The advertising exposure of the DTCI campaign conducted was for 5 weeks (November 19, 2011 to December 22, 2011) throughout Japan via television, Internet, and print advertising [8]. During a 30-s television commercial that aired from November 19, 2011 to December 11, 2011, a well-known Japanese celebrity introduced the symptoms of overactive bladder. In the television commercial, a doctor also suggested, quote, “Overactive bladder could be cured by pharmaceutical treatment.” and “People with related symptoms should inform their healthcare providers.”
We assumed that the study population was equally exposed to the advertising campaign because it aired throughout Japan [8]. To the best of our knowledge, there was no other plausible factor to affect prescription patterns (such as a guideline update or another DTCI campaign/disease awareness campaign/public awareness campaign) during the follow-up period, except for the release of mirabegron to the pharmaceutical market on September 16, 2011 prior to the DTCI campaign.
Main outcome, observation period, and covariates
The main outcome was first-time prescription of any treatment drug during the 3-year analysis period (November 19, 2010 to November 18, 2013) to capture the impact of the DTCI campaign on treatment drug prescription.
We assessed the impact of the DTCI campaign using a survival analysis. Aligned with the campaign start date on November 19, 2011, we divided the 3-year analysis period into three yearly periods: the Pre-Campaign Year (November 19, 2010 to November 18, 2011), Year 1 (November 19, 2011 to November 18, 2012), and Year 2 (November 19, 2012 to November 18, 2013). Because the campaign duration was 5 weeks, we created 10 observation periods of 5 weeks’ length each, starting on November 19 in each year: Period 1 (weeks 1–5), Period 2 (weeks 6–10), and consecutively to Period 10 (weeks 46–50). The period after week 50 was excluded in each year. The campaign was conducted during Period 1 in Year 1. Patients in Year 1 and Year 2 were exposed to the advertising campaign. Patients in each period had not been prescribed any treatment drugs for overactive bladder at the beginning of the period. The observation was right-censored in each period.
Covariates included 29 dummy variables indicating these 10 periods for 3 years (reference = Period 10 in the Pre-Campaign Year). Additional covariates were sex, age, and comorbidity levels. Age and comorbidity were time-varying variables and could be assessed only at the beginning of each year due to the limited data availability. The Charlson Comorbidity Index with ICD-10 was used for assessing comorbidity levels [12].
Statistical analysis
First, to detect the change in prescription patterns owing to the DTCI campaign, Kaplan–Meier estimates were made for first-time prescriptions as the prescription rate during the overall 3-year analysis period.
Second, we ran a Cox proportional hazard model to investigate the magnitude and continuity of changes in prescription patterns as a result of the DTCI campaign. By comparing with Period 10 in the Pre-Campaign Year, we estimated hazard ratios (HR) and 95% confidence intervals (CI) for each of the remaining 29 periods as indicators of the prescription rate ratio, adjusted for sex, age, and comorbidity levels. The equation used for analysis was as follows:
$$ {\displaystyle \begin{array}{l}\left(\mathrm{First}\ \mathrm{time}\ \mathrm{prescription}\right)={\beta_{11}}^{\ast}\left(\mathrm{Female}\right)+{\beta_{12}}^{\ast}\left(\mathrm{Age}\right)+{\beta_{13}}^{\ast}\left(\mathrm{Comorbidity}\right)\\ {}\kern10em +{\beta_{21}}^{\ast}\left({\mathrm{P}}_1\ \mathrm{in}\ {\mathrm{Y}}_{\mathrm{pre}}\right)+\cdots +{\beta_{29}}^{\ast}\left({\mathrm{P}}_9\ \mathrm{in}\ {\mathrm{Y}}_{\mathrm{pre}}\right)\\ {}\kern10em +{\beta_{31}}^{\ast}\left({\mathrm{P}}_1\ \mathrm{in}\ {\mathrm{Y}}_1\right)+\cdots +{\beta_{40}}^{\ast}\left({\mathrm{P}}_{10}\ \mathrm{in}\ {\mathrm{Y}}_1\right)\\ {}\kern10em +{\beta_{41}}^{\ast}\left({\mathrm{P}}_1\ \mathrm{in}\ {\mathrm{Y}}_2\right)+\cdots +{\beta_{50}}^{\ast}\left({\mathrm{P}}_{10}\ \mathrm{in}\ {\mathrm{Y}}_2\right)\\ {}\kern10em +\mathrm{residual},\end{array}} $$
where Y represents Year (subscript is Pre (-Campaign Year), 1 or 2), and P for Period (subscript ranges from 1 to 10). In the equation above, Period 10 in the Pre-Campaign Year was the reference period. Baseline seasonal trends were evaluated using variables of the Pre-Campaign Year. In addition, we calculated crude prescription rate ratios per 100,000 person-days for periods with significant HRs as compared with the Pre-Campaign Year, to validate the magnitude of HRs.
To confirm the effect of DTCI campaign at the population-level, we performed interrupted time series analysis (ITSA) [13, 14] with aggregated data of the first-time prescription rate per each period (5-week) per standardized 100,000 persons (N1 of analyzed time periods = 30). We ran user-written STATA command of “itsa” with Prais-Winsten and Cochrane-Orcutt regression [13, 14]. We set the interrupted time point of the intervention at Period 4 in Year 1 in our primary ITSA analysis. Our primary ITSA analysis assumed a three-month potential time lag partly because an average interval of clinic visits was suggested to be 3 months among overactive bladder patients in Japan by a previous study [15] and also partly because it usually takes up to 3 months for overactive bladder patients to receive the first-time time treatment drug prescription when their urologists follow the Japan’s clinical guidelines that recommend other clinical procedures (i.e., screening of other possible underlying diseases, fluid intakes assessment, and pelvic-muscle floor training that may take up to 3 months in in a regular clinical setting in Japan) prior to the prescription [11, 16].
Additionally, to assess potential different time lag regarding the impact of the DTCI campaign, we performed sensitivity analyses of the ITSA with various interrupted time periods: from Period 1 to Period 5 in Year 1 (Period 5 corresponds to a 20-week delayed time lag) besides the primary ITSA analysis.
The equation used for the ITSA was as follows [13, 14]:
$$ {\displaystyle \begin{array}{l}\left(\mathrm{Aggregated}\ \mathrm{outcome}\right)={\beta}_0+{\beta_1}^{\ast}\left(\mathrm{time}\ \mathrm{since}\ \mathrm{the}\ \mathrm{start}\ \mathrm{of}\ \mathrm{the}\ \mathrm{study}\right)+{\beta_2}^{\ast}\left(\mathrm{post}-\mathrm{intervention}\right)+\\ {}\ {\beta_3}^{\ast}\left(\mathrm{time}\ \mathrm{since}\ \mathrm{the}\ \mathrm{start}\ \mathrm{of}\ \mathrm{the}\ \mathrm{study}\right)\times \left(\mathrm{post}-\mathrm{intervention}\right),\end{array}} $$
where
-
1.
β0 is an “intercept” in a regression, representing the starting level of the outcome variable.
-
2.
β1 is a “slope” in a regression and indicates trajectory of the outcome variable prior to the introduction of the intervention. If β1 is not statistically significant, the outcome level remains constant at β0 prior to the intervention. In this case, β0 also represents the level of the outcome variable immediately before the intervention.
-
3.
β2 represents the “one-time” change in a “regression-intercept” or the level of the outcome that occurs in the period immediately after the intervention, which is hypothesized to be caused by the intervention.
-
4.
β3 represents the “long-term” change, expressed as the “slope” difference between pre-intervention and post-intervention in a regression, which is also hypothesized to be caused by the intervention.
Time since the start of the study is a continuous variable, and post-intervention is a dummy variable (post-interrupted time point, 1; otherwise, 0).
For supplementary data analysis with a limited monthly aggregated data set extracted from the whole JMDC cohort during November 2010 to November 2012 (N2 of analyzed time periods = 25), we also checked the changes on different but relevant outcomes (a) the number of new diagnosis and (b) the number of newly diagnosed patients treated with medication.
All p-values were two-sided, and p < .05 was considered statistically significant. Data were analyzed using STATA/MP13.1 (StataCorp LP, College Station, TX).