Data were from two published 13-week, randomized, placebo-controlled studies in male and female outpatients who were at least 40 years of age, and met American College of Rheumatology clinical and radiographic criteria for the diagnosis of OA of the knee [15, 16]. In both studies, pain was assessed daily using an ordinal 11-point numerical rating scale (0 to 10) that was recorded in patient diaries. Efficacy assessments were based on the weekly mean of the daily average pain severity from these patient diaries. Pain was also assessed with the Brief Pain Inventory (BPI)  (severity range: 0–10) at weeks 4, 7, and 13. All patients provided written informed consent before the commencement of any study procedures.
For both studies, patients were required to have pain for ≥14 days of each month for 3 months prior to study entry, with a mean pain severity ≥4 on daily pain diary ratings during the screening and baseline periods. In addition, patients had to agree to maintain their usual activity level throughout the course of the study. Key exclusion criteria for both studies included having a body mass index >40 kg/m2; a diagnosis of inflammatory arthritis or an autoimmune disorder; having received invasive therapies within the prior 3 months, or joint replacement to the knee; being non-ambulatory or needing assistance walking with crutches or walker; any serious medical condition or psychiatric disorder, including major depressive disorder, as identified by the Mini International Neuropsychiatric Interview  that could compromise participation in the study.
Patients were randomly assigned 1:1 in double-blind fashion to treatment with duloxetine 60 mg/day or placebo as determined by a computer-generated random sequence using an Interactive Voice Response System (IVRS). All patients randomized to duloxetine were started on 30 mg/day for one week then escalated to 60 mg/day. After 7 weeks, the duloxetine dosing regimen could change. In Study I , patients in the duloxetine group were re-randomized via IVRS without regards to change in pain severity to either continue duloxetine 60 mg/day or to have their dose increased directly to 120 mg/day, and were continued on that dose for the remainder of the study. In Study II , patients in the duloxetine group who had <30% reduction from baseline in pain severity on the BPI 24-h average pain severity scale, had their dose increased directly to 120 mg/day and continued on that dose for the remainder of the study.
For this post hoc analysis, two patient groups were defined by their age at study entry: older patients were at least 65 years of age, and younger patients were less than 65. For the efficacy and safety analyses comparing duloxetine with placebo within age groups, studies were pooled and doses were pooled. In addition, a subgroup analysis was conducted to determine the efficacy of duloxetine in the “oldest of the old” (>75 years) as compared with the “younger of the old” (65 to <75 years) patients. For comparing the effect of increasing duloxetine to 120 mg in non-responding patients, only data from Study II were evaluated as it most closely reflected clinical practice; whereas re-randomization to a higher dose does not. Treatment arms were defined post hoc as placebo, duloxetine 60 mg/day (patients who remained on 60 mg for the entire study), and duloxetine 60/120 mg/day (patients who received 60 mg for 7 weeks, followed by duloxetine 120 mg for 6 weeks). Safety assessments included spontaneously reported treatment-emergent adverse events (TEAEs), discontinuation rates due to TEAEs, serious adverse events (SAEs), and treatment-emergent abnormal vital signs.
Categorical baseline characteristics were compared between age groups using a logistic regression model with terms for study, treatment, age group and the treatment-by-age group interaction. Quantitative baseline characteristics were compared using an analysis of variance model with similar explanatory terms. Weekly means of change from baseline in the average pain severity ratings from patient diaries were analyzed using a likelihood-based, mixed-effects model repeated measures (MMRM) approach that used all available observations. The model included the fixed categorical effects of site, treatment, week, age group, treatment-by-week interaction, age group-by-week interaction, treatment-by-age group interaction, treatment-by-age-group-by-week interaction, as well as the continuous fixed covariates of baseline score and baseline-by-week interaction.
The analysis of completion rates, discontinuation rates due to adverse events (AEs), and incidence of spontaneously reported TEAEs and SAEs were compared using a logistic regression model with terms for study, treatment, age category, and the treatment-by-age interaction. For comparison among treatment groups and age groups, test results were considered statistically significant if p≤.05; for tests of interaction, results were considered statistically significant if p≤.1. Statistical analyses were performed using SAS, version 9.1 (SAS Institute, Inc, Cary, NC).