1. Describe the study design you will employ in order to determine if your intervention has had an effect on the outcome variable of interest.
The first stage of my proposed project will be a pilot study of continuous labor support services (doula services) at the UCSF Medical Center Labor & Delivery ward. As only one site will be involved at this stage, I plan to use an interrupted time series design. The aim of the study is to measure change in the proportion of births delivered by c-section before as compared to after implementation of the intervention. As there are well documented time trends in the rate of cesarean sections (they have been increasing, although possibly starting to drop due to recent/growing backlash), I will want to include at least one year of weekly or monthly data on proportion of births delivered by cesarean section leading up to intervention implementation, followed by weekly/monthly data as the program is implemented, and capped by at least one year of weekly/monthly data following full implementation of the intervention. In addition to proportion of births delivered by c-section, I will also track secondary indicators such as first c-section, instrumentation used in delivery (forceps, vacuum), pain medications used (including epidural), length of labor, and patient reported satisfaction with labor.
As a further attempt to account for any temporal trends that may occur over the study period, I would attempt to collect these same weekly/monthly measures from 2-3 other hospitals in the Bay Area that are deemed to be comparable on important characteristics, to the extent possible. We would then have this control data to use as a comparison to any trends observed following the intervention at UCSF.
In terms of statistical analysis, I need to invest more time in learning about the options in more detail, but it seems likely that I might used either ARIMA or xtmixed to account for the autocorrelated nature of the data points over time.
2. Define the unit-of-analysis for your main outcome evaluation, the minimum meaningful effect size, and the sample size necessary to detect this effect size.
The unit of analysis is each individual birth at UCSF – specifically, measuring whether it each infant was delivered vaginally or by c-section. These individual birth level data will be aggregated to form a proportion of births delivered by c-section either by week or by month.
An estimate for the minimum meaningful effect size, conservatively estimated given what was identified in a recent Cochrane Review, might be a decrease in the absolute proportion of births delivered by c-section by 5 percent. (The Cochrane Review found that continuous labor support resulted in a 22% reduced risk of c-section as compared to laboring mothers not provided continuous labor support). In other words, I would hope that this intervention would result in a reduction in the average percentage of births delivered by c-section by at least 5 percent. Given that a reported 23 percent of births at UCSF are now delivered by c-section, a reduction of 5 percent would translate to an observed c-section rate of 18 percent. To be able to detect an effect of this size (5% reduction), many time points may be required.
The elements needed to calculate a required sample size include an estimate of the hypothesized effect size (5%), the standard error of the effect size (roughly 4%, loosely estimated from Cochrane Review findings), the desired alpha level (0.05), and an inflation factor to account for correlation among included covariates (an estimate of 1.3), and the autocorrelated nature of the data (we will assume a moderate degree of autocorrelation, 0.25)
We know that UCSF Parnassus delivered 1,858 infants in 2013. This translates to a monthly figure of about 155 deliveries. I cannot “increase” the sample size in terms of number of deliveries performed, but I can increase the sample size to include more time points collected before and after the intervention to better account for time trends, and to increase the precision of the estimates. Thus, assuming I can access the data, I will include 24 time points prior to the intervention (monthly for the prior two years), as well as up to 12-24 months following the intervention. As mentioned above, I will also attempt to get c-section data from another hospital in the Bay Area (where a doula intervention is not in place) to compare and better adjust for time trends. I have not yet been able to figure out how to calculate the power I would have to detect this 5% decrease with this number of time points, but I am working on it!