Effect of enhanced feedback to hospitals that are part of an emerging clinical information network on uptake of revised childhood pneumonia treatment policy: study protocol for a cluster randomized trial

The overall aim of the trial described in this protocol is to assess the effect of enhancements to feedback on clinician adoption of new pneumonia treatment guidance within county hospitals that are already receiving regular standardized audit reports spanning several clinical conditions. During the study we will examine one primary objective and a secondary objective:

We designed a parallel-group, pragmatic, cluster randomized trial to be conducted in 12 hospitals over a 9-month period with interim analysis at 6 months in case of a more marked intervention effect than we are expecting. At the time of the trial design all the 12 hospitals were participating in a clinical network in which standardized audit and feedback had been delivered to all sites for more than 12 months; therefore, providing context for testing an enhanced feedback intervention. The intervention was designed to examine whether specific characteristics of feedback (frequency, goal setting, and action planning) can result in greater impact (Fig. 2). The intervention targets improvements in hospital uptake of new pneumonia treatment policy; therefore, hospitals (and entire pediatric care teams) rather than individual clinicians will be randomized to intervention [15].

County referral hospitals form clusters representing diverse geographic regions with varying climatic conditions, morbidity patterns, and facility workload. Out of the 14 hospitals currently engaged in network activities, two hospitals will not participate in the trial because one site located within the capital, Nairobi, serves as a satellite of the main university teaching hospital while the second hospital is a small rural training facility co-located with a larger facility participating in the network. The rural training facility has a different type of patient population and no resident physician. The CIN was set up between February and October 2013 through recruitment of hospitals, based on preliminary consent, from a shortlist of 40 public county referral hospitals providing internship training at the time and serving as first-referral centers. Hospital selection and characteristics are described in more detail elsewhere [16]. The exclusion criteria for hospitals from the eligible population included small hospitals (fewer than 1000 annual pediatric admissions), and existing formal linkages with research or academic institutions working with hospitals to conduct research. The patient population eligible for inclusion within each hospital was pediatric patients admitted with a pneumonia diagnosis in a child with no indication for broad-spectrum antibiotic therapy. Common indications for broad-spectrum antibiotics that will define the exclusion criteria include severe acute malnutrition, and meningitis. Severe malaria and tuberculosis cases will also be excluded. All hospitals recruited in the trial assented to participate and were informed that they could discontinue participation in the trial without prejudicing their role within the CIN.

The CIN comprises a multifaceted approach including audit and feedback for all hospitals to which specifically enhanced feedback on the uptake of new pneumonia treatment policy is added for intervention sites.

The new childhood pneumonia treatment policy advocates for assessment of respiratory rate, lower chest wall indrawing and five specific danger signs (cyanosis, oxygen saturation below 90%, inability to drink/breastfeed, altered level of consciousness, and grunting) in children presenting with cough or difficulty breathing followed by classification into one of two severity classifications: pneumonia (for children with raised respiratory rate and/or chest wall indrawing) or severe pneumonia (for those with any of the remaining danger signs). The guidelines further recommend treatment with orally administered amoxicillin for pneumonia and intravenously administered antibiotics for severe pneumonia [12]. A composite indicator for the classification and treatment of pneumonia was used for the primary outcome. The indicator was developed by reconstructing a pneumonia classification for each patient admitted with cough or difficulty breathing using information documented in medical records on whether each of the six specific clinical signs required for pneumonia severity classification were present or absent (one network effect has much better clinical documentation). This classification, based on a clinical guideline algorithm, was then compared to the classification assigned by the admitting clinician to determine whether the clinician classification was correct. The children with a correct pneumonia classification were then evaluated to determine whether they received orally administered amoxicillin, the recommended treatment for pneumonia, yielding a composite indicator of correct classification and treatment.

The trial will be conducted in 12 geographically dispersed hospitals participating in the CIN. Restricted randomization was used to allocate the participating hospitals into the intervention and control arms to ensure that balance is achieved between the two treatment arms with respect to number of patients admitted as inpatients with pneumonia, and geographical location (also a proxy for malaria prevalence or disease case-mix) [15]. The hospitals were grouped into four categories using information on geographic location (central or western region) and monthly average number of patients admitted with pneumonia (<30 or ≥ 30 per month) collected within the clinical network during the year preceding the intervention. Five hospitals were located in the western region, while seven hospitals had at least 30 patients admitted with pneumonia per month. A statistician used R statistical software to generate the 924 possible allocations for assigning 12 clusters to two treatments with six clusters allocated to each treatment arm. We defined a balancing criteria based on the two selected covariates and determined that approximate balance would be achieved by including allocations which had in each treatment arm either two or three hospitals located in the western region and three or four hospitals admitting < 30 pneumonia cases per month. The R computer program was used to assess each allocation against the balancing criteria for geographic location (2:3 or 3:2 split between the arms) and monthly number of patients admitted with pneumonia (3:4 or 4:3). There were a total of 536 allocations that met the criteria and represented relative balance between the intervention and control arms with respect to desired allocation ratio. Computer-generated random numbers were used to select one of these allocations at random. Hospitals in the randomly selected allocation were assigned to treatment and those not included in the allocation were assigned to the control group. Allocation was based on clusters rather than individual patients and allocation concealment was conducted at the cluster level. The trial statistician concealed the sequence from investigators until training on the new policy was conducted in all hospitals and the intervention was assigned. Blinding of investigators and health care providers after intervention assignment was not feasible in the trial because the delivery of enhanced feedback on pneumonia care provided was done by investigators and the target of this feedback was the health care providers. However, for the interim analysis the statistician was blinded to treatment arm allocation.

Based on the primary outcome of correct pneumonia classification and treatment we performed a power calculation to determine the feasibility of conducting the trial within the period of the network’s funding cycle using a fixed number of hospitals recruited into CIN. Using data collected during the first year of CIN operation we determined that: on average, 616 (range 310–898) children are admitted to each hospital annually with pneumonia and of these 150 children (range 78–228) are eligible for treatment with amoxicillin. We estimated that if we evaluate the primary outcome 9 months after trial commencement with six hospitals per arm then at least 680 patients with pneumonia requiring amoxicillin treatment will be enrolled per arm. Assuming an average cluster size of 113 patients and an intracluster correlation coefficient (ICC) of 0.2 (corresponding to a variance inflation factor (VIF) of 15.8 and based on calculation of pneumonia classification and treatment within CIN hospitals), a Type-I error rate of 5%, and 90% power we will be able to detect a two-fold increase in the odds of correct pneumonia classification and treatment over 9 months if correct pneumonia classification and treatment is 30% in the control group following the treatment policy change. To ensure that the desired sample size is attained within each hospital, all eligible patients will be recruited in the hospitals with a workload less than 3000 annual pediatric admissions. For hospitals with a very high annual workload (at least 4000 admissions) a random sample will be selected using computer-generated random numbers applied to eligible participants.

The primary outcome is correct patient assessment and treatment-plan-defining care in the first 24 h. This endpoint was evaluated as the proportion of all patients admitted with pneumonia who were eligible for treatment with orally administered amoxicillin based on guideline recommendations (increased age-specific respiratory rate or lower chest wall indrawing in a child with history of cough or difficulty breathing) who are correctly classified and treated using orally administered amoxicillin at 24 h. The secondary outcome will be measured over the course of the admission as any change in treatment for pneumonia after the first 24 h. This change from orally administered amoxicillin to intravenously administered antibiotics for pneumonia during inpatient stay is considered a proxy for treatment failure rate..

Scientific and ethical clearance to conduct the trial was obtained from the Scientific and Ethics Review Unit of the Kenya Medical Research Institute. The approval is updated annually and provided for the establishment of the CIN and the conduct of this specific trial within it, and provides for the use of de-identified patient data, obtained through retrospective review of medical records without individual informed consent, for the purposes of service evaluation and implementation research.

Ongoing.