A written action plan, by our definition, had two components: an algorithm that identified specific clinical indicators signaling the need for adjustments in medication; and specific instructions on how to adjust medications in response to such indicators. Many publications lacked sufficient detail on the written plan, so a brief survey was sent to the primary author of each of the 36 studies. If no response was obtained (36%), the article was excluded only when it was clear from the publication that our definition was not met.
Assessment of study quality
High-quality studies were randomized controlled trials that met the 3 domains of study quality that have been demonstrated empirically to impact effect size: concealment of treatment allocation; double-blinding; and minimization of exclusion bias.9,10 However, we doubted the feasibility of double-blinding a written asthma plan intervention, and so relaxed this requirement. We considered exclusion bias to be minimized when a study either reported intent-to-treat analysis or excluded fewer than 10% of subjects from analysis, with the ratio of subjects excluded from each arm being less than 2:1.
To more fully evaluate study design issues that may be particularly important in asthma research,11,12 we constructed asthma-specific quality indicators in consultation with an expert panel. Controls for potential confounders of treatment effect included establishing reversibility of airway obstruction, controlling for other medication use, reporting compliance, and addressing seasonality. In addition, a priori reporting of power calculations and accounting for exclusions and withdrawals were judged to be study quality characteristics pertinent to this body of evidence.
Data analysis
We constructed evidence tables for the outcomes of interest, and performed a qualitative synthesis of the data. Meta-analysis was not appropriate due to wide discrepancies in the patient populations studied, the interventions employed, and measurement and reporting of outcomes.
Results
Our literature search yielded a total of 4578 citations. Of these, 36 studies met the initial selection criteria. Many of these qualifying studies, however, were confounded by multiple asthma management interventions applied inconsistently across treatment arms. For example, a common confounder was review of and change in long-term medication use in the treatment group, but not in the control group. This necessitated a refinement in our selection criteria to focus on studies that largely isolated the effect of written action plans.13-21 This step yielded a final evidence base of 9 randomized controlled trials with a total enrollment of 1501 patients.
Table 1 summarizes the characteristics, interventions, and outcomes of the 9 studies. Two studies were 3-arm trials,16,17 which raised the total number of comparisons among the 9 studies to 11. The largest study was the Grampian Asthma Study of Integrated Care (n=569),14 a community study conducted in the UK. Enrollment in the other 8 studies ranged from 43 to 64 patients per arm. Treatment duration ranged from 24 to 52 weeks.
None of the studies met our definition of high quality. In fact, no study met any of the generic quality criteria—none was blinded, none described concealment of allocation, and all excluded more than 10% of subjects. Furthermore, none reported an intention-to-treat analysis. Thus these trials were prone to withdrawal bias as well as overestimation of treatment effect due to lack of allocation concealment.
No study met the majority of asthma-specific indicators (Table 1). Of the 9 studies, only 5 met any asthma-specific indicator. Three reported prospective power calculations,13-15 but 2 of these substantially overestimated the expected effect.13,15 Two studies established reversibility;14,17 2 controlled for other medication use;13,15 and 2 reported compliance.17,21 Thus, the studies were also prone to a type II error (failing to detect a true effect) and to potential confounding of outcomes.
We performed sample power calculations for hospitalizations (Table 2), derived from baseline rates reported in 4 studies14,16-18 and standard deviations reported in 2.14,17 A study with 250 patients per arm could detect a reduction of 50% or more in hospitalization, given a control rate of at least 0.2 hospitalizations/patient/year. In actuality, GRASSIC,14 which is the largest available trial (N=569), had baseline hospitalization rates of 0.12 and 0.13. With this baseline rate, over 700 patients per arm are required, higher than the actual enrollment in GRASSIC. The other studies in this review would be adequately powered to detect a 50% difference only in the setting of even higher baseline utilization (eg, 0.30 hospitalizations/patient/year).
Table 3 displays utilization outcomes for the 11 comparisons in the 9 trials. In 5 studies (N=1019), medical management with a written action plan was compared with medical management without a written action plan.13-17 Two trials (N=185) compared a peak flow meter plus a written action plan with a peak flow meter and no written action plan18,19 In 4 studies (N=393), a written action plan based on peak flow monitoring was compared with a written action plan based on symptoms.