488Trial design and statistical considerations for a trial comparing group and individual treatments

abstract

Abstract Focus of Presentation A randomized clinical trial is considered the best experimental study design for comparing the effect of an intervention against a control. Selecting a trial design depends on the trial objective, a better choice would be a design that provides optimal estimation of the intervention effect, i.e. a design that yields smaller variance for the estimated intervention effect and giving stable estimates. The choice of analysis method depends on trial design aspects, reasonable assumption on the underlying probabilistic model generating the data and the credibility of findings depend on the appropriateness of the analysis method used. Here, we describe a planned randomized standard of care-controlled trial on interactive weekly mobile text messaging added to a motivational interviewing intervention aimed at sustaining continued breastfeeding among women living with HIV in South Africa. Under the “findings” heading, we highlight some of the trial design and statistical issues for discussion at the early career workshop, to gain insights on possible approaches to address the methods issues. Methods and trial design: Women from peri-urban informal settlements and a rural setting will be invited to participate within 24 hours of giving birth at selected primary healthcare facilities. Eligible women will be individually randomized to intervention or control arm after providing written informed consent. The intervention will consist of a weekly text message encouraging women to exclusively breastfeed and inquiring if they have any problems breastfeeding their infants. In addition to text messaging, women assigned to the intervention arm must visit the research site where a research nurse or counsellor will conduct an individual motivational interview, at weeks 2, 6, and 10 post-delivery. Women assigned to the control arm will be counselled by the standard of care service, i.e. primary healthcare nurses and trained counsellors will counsel women to exclusively breastfeed for the first six months through group educational infant feeding counselling sessions. Study participation will not change standard of care of participants, so women assigned to the intervention arm will receive the standard of care service, in line with provincial guidelines applicable in the sector during the study period. The primary outcomes are 1) number of women who are exclusively breastfeeding at week 24 post-delivery and 2) number of women reporting any breastfeeding at week 24 post-delivery. Findings or trial design and statistical issues Although women are individually randomly assigned to intervention and control arms, the standard of care at the primary health care facility may induce a dependency between these participants, also called the group therapy effect. For example, promotion of exclusive breastfeeding in HIV-infected women may be enhanced through positive feedback from women who follow this practice or vice versa. The standard of care will be clustered within group infant feeding counselling session, where participants (both assigned to the control and intervention arm) will receive the standard of care in groups. Women assigned to the intervention arm will receive individual motivational interview and the text messages will be send separately, to each woman’s mobile phone. Implications The trial objective will be to determine whether at week 24 following delivery, weekly text message added to motivational interview leads to better adherence to exclusive breastfeeding and leads to extended breastfeeding than standard of care. Trial design and statistical issues to discuss during the workshop will include a discussion around: 1) the appropriateness of invoking the standard experimental design to address the study objectives, and possible alternatives 2) Sample size estimation for a trial comparing group and individual treatments? 3) Reasonable assumption on the underlying probabilistic model generating the outcome data and 4) Candidate statistical models. Key messages Use of sound statistical principles of experimental design established for clinical trials allow objective and unbiased comparisons.