The methods used to recruit community pharmacists, family physicians and their elderly patients have been described elsewhere.23
Briefly, a convenience sample of 24 pharmacists who had received additional post-university training in the prevention, identification and resolution of drug-related problems24
was approached in 16 towns or cities in southern Ontario. All family physicians who practised in each pharmacist's postal code area made up the sampling frame. A random sample of physicians in each postal code area was generated, and physicians were approached by telephone until 2 (1 pair) had been recruited in each area (). About 20 randomly chosen eligible senior citizens per practice (cluster) were recruited (range 7–23 per practice) by the office staff of the practice from August to November 1999. Patients were eligible for inclusion in our study if they were aged 65 years or more, taking 5 or more medications, had been seen by their physician within the past 12 months, had no evidence of cognitive impairment and could understand English. Patients were excluded if they had planned surgery, were on a nursing home waiting list or were receiving palliative care. The study was approved by the Research Ethics Board of Hamilton Health Sciences.
Fig. 1: Flow diagram showing the recruitment of family physicians and patients into the randomized controlled trial. R = randomization.
Before the physicians were randomly allocated, 1 of 8 specially trained research nurses assigned to each practice administered questionnaires designed to collect data on sociodemographic characteristics, medication use and quality of life from the study patients. A list of current and past medical conditions was compiled by the nurse for each patient and confirmed by his or her physician. An ICD-9 (International Classification of Diseases
, ninth revision)25
code was assigned for each diagnosis and reviewed by a family physician (J.S.).
The pair of physicians in each postal code area were randomly allocated, in a concealed fashion, to the intervention or control group, using a central telephone randomization procedure based on computer-generated random numbers (). Randomization was conducted by a research team member (J.K.) who was blinded to the practices' identities. Study patients in practices allocated to the control group received usual care from their physicians. Neither family physicians nor their patients were blinded to their allocation group.
A study patient in a practice randomly allocated to the intervention group had a structured medication assessment by the pharmacist in the physician's office. After the interview, the pharmacist wrote a consultation letter to the physician that summarized the patient's medications, identified drug- related problems and recommended actions to resolve any such problems. The pharmacist subsequently met with the physician to discuss the consultation letter. After the meetings, physicians used a data collection form to indicate which recommendations they intended to implement and when. The pharmacist and physician met again 3 months later to discuss progress in implementing the recommendations. Five months after the initial visit, the pharmacist met with the physician to determine which recommendations had been put in place. One and 3 months after meeting with the physician, the pharmacist monitored each patient's drug therapy using a semistructured telephone interview with the patient.
A drug-related problem is defined in pharmaceutical care as any undesirable event experienced by a patient that involved or was suspected of involving drug therapy and that actually or potentially interfered with a desired patient outcome.24
Eight categories of drug-related problems () are widely used by pharmacists to administer patient-centred care.18,22,26
The intervention model to identify, resolve and prevent drug-related problems was deemed to be feasible and acceptable for implementation in primary care based on a pilot study with input from family physicians and pharmacists.27
Drug-related problems were determined by the pharmacist consultants using information from the patients' medical charts, the face-to-face interviews and the medication reviews, and they were recorded on a standardized data form.
Research nurses interviewed the patients in both groups and recorded their current use of prescribed and over-the-counter medications. The nurses returned to the same practices and requested that the patients not mention whether they had met with a pharmacist. However, the extent to which the research group allocation became known to the nurses was not formally assessed.
The primary end-point measure was a reduction in the daily units of medication taken, as a surrogate for optimized drug therapy. A unit was defined as 1 tablet, 1 teaspoon, 1 drop (eye), 1 application of cream or ointment, or 1 dose of insulin. Other short-term outcome measures that were thought to reflect optimized drug therapy included the costs of medications, health services use, and health-related costs and quality of life.
The number of units and costs of daily prescription and over-the-counter medications were determined for each patient at the beginning and end of the study. Ontario Drug Benefit Program prices were used as the cost of the drugs covered under this plan. All other prices were obtained from a commercial drug wholesaler database or from local pharmacies, if absent from the database. Average daily costs were calculated for all medications.
Information was gathered on the use of health services during the study period from the patients' medical charts and from diaries completed by the patients for health services that would not normally be in the medical charts. Fees for physician services were obtained from the Ontario Schedule of Benefits for Insured Medical Services. The cost of hospital stays and other health services costs were obtained from an area hospital that was participating in the Ontario Case Costing Project (www.occp.com
). A hospital stay was considered to be medication-related if it had “probably” or “definitely” resulted from the effects of a medication or medications on a patient's health. To separate hospital stays caused by medication problems from other hospital stays, 2 experts who were unaware of study subjects' allocation (L.D. and a physician) independently assessed each hospital stay as medication-related or not, using a list of the senior's medications and medical conditions at baseline, the reason for admission and the death certificate, if applicable. Any disagreements were resolved by discussion. The Medical Outcomes Study 36-item Short Form (SF-36)28
quality-of-life survey was self-administered at the enrolment and exit interviews.
To evaluate the implementation process within the pharmacist–physician pairs, we also determined physician perception of the service and the extent to which physicians agreed with the recommendations made by the pharmacists for the patients in the intervention arm. In addition, each pharmacist–physician pair jointly assessed whether each recommendation had been implemented (fully or partially) or attempted.
The experimental unit (and the unit of analysis) was the family physician, and the patients were considered to be nested within each physician's practice.29
The desired sample size of 48 physicians with 15 patients per physician was estimated using an intracluster correlation coefficient of 0.08 for daily units of medication, based on a pilot study27
and the desire to detect a 15% reduction in daily units of medications in the intervention arm, as compared with the control arm. The hypothesized effect size was based on the results of the pilot study27
and was felt to represent a clinically important reduction in the number of daily medication units. To account for the design, both sample size calculations and the analysis were based on a random effects meta-analysis across cluster pairs, proposed by Thompson and colleagues.30
In this method, the mean differences in the outcome variables between groups are weighted averages of the mean differences across clusters, or practices, and are compared using an asymptotic χ2
test. A one-tailed type I error (alpha) of 0.05 was used for all statistical tests, with a power of 80% to detect the difference.