Strengths of the MANA Stats 2.0 Dataset
The MANA Stats 2.0 dataset includes data on complete courses of prenatal, birth, and postpartum care for 20,893 pregnant women who planned a midwife-led home or birth center birth. These data have some unique properties that will allow for the examination of research questions related to midwife-led births, normal physiologic birth, and birth outcomes by intended place of birth. In 2009, 62% of home births in the United States were attended by midwives: 19% by CNMs and 43% by “other midwives,” including CPMs and other direct-entry midwives. However, CPM practice and outcomes are understudied. More than 70% of births in the MANA Stats 2.0 sample were attended by CPMs, allowing researchers a unique opportunity to study practices and outcomes associated with these providers.
In addition, interventions such as continuous electronic fetal monitoring, labor induction and augmentation with synthetic oxytocin, and epidural anesthesia are so ubiquitous in contemporary hospital settings[12, 40] that it is challenging to find perinatal datasets that contain a large number of undisturbed births. The AABC's UDS and MANA Stats share a similar capacity for the collection of data on normal physiologic birth, and both include home and birth center births. However, the UDS draws data primarily from birth centers, while the MANA Stats 2.0 dataset predominantly contains courses of care for women who intended to give birth at home.
Most important, this dataset is valuable to researchers because it allows for the analysis of outcomes by intended place of birth. As has been noted in studies examining trends in birth setting in the United States,[38, 41] it is currently impossible to reliably study outcomes by place of birth using vital statistics data alone because, to date, in most states, birth certificates collect only actual place of birth and not intended place of birth. Analysis of birth outcomes by setting are subject to misclassification bias because some intended home and birth center births actually occur in the hospital following intrapartum transfer, just as some planned hospital births accidentally occur at home in the absence of a professional birth attendant. Reliable evaluation of safety and efficacy of midwife-led birth across birth settings can only occur when women are correctly classified according to intended place of birth at the onset of labor.
Pre- and Postreview Analysis
An analysis of selected variables pre- and postreview revealed kappas ranging from 0.98 to 1.00, where kappa greater than 0.7 is considered good. A 2012 systematic review of the quality of data in perinatal population health databases summarized sensitivity, specificity, and kappa ranges for 43 studies that compared perinatal data collected for research purposes against a gold standard, usually medical records. While some data points (eg, mode of birth) had excellent agreement, many other variables showed much lower kappa ranges, revealing errors and inaccuracies in perinatal data collection. In comparison, the MANA Stats 2.0 data were accurately entered by participants, as evidenced by the perfect or near perfect agreement among pre- and postreview variables. This suggests that any errors in this 2.0 dataset are primarily random and not systematic, at least for the key outcomes assessed.
There are several limitations of the MANA Stats 2.0 dataset for the purposes of research. The primary one is that the sample was captured through voluntary participation by providers; thus, it may not accurately reflect population-based outcomes. In addition, we are unable to quantify precisely what proportion of practicing midwives contributed data between 2004 and 2009 for a number of reasons. For example, a total of 54 CNMs contributed data between 2004 and 2009. According to the National Center for the Analysis of Healthcare Data, there were 7922 CNMs licensed in the United States in 2009. Given that the known proportion of CNMs attending home births is less than 4%, we estimate that approximately 316 CNMs were attending home births in 2009, for a participation rate of about 17% for MANA Stats. In addition, only 12.5% of births in the MANA Stats 2.0 dataset were attended by CNMs. As such, this dataset captures a very small proportion of the total births attended by CNMs; thus, it cannot be used to reliably describe CNM practice outcomes—or even those outcomes for the subset of CNMs who attend home and birth center births.
Although a much larger proportion of the total number of active CPMs contributed data between 2004 and 2009, there are still barriers to estimating a participation rate for the 6-year period. The number of CPMs in the United States has increased sharply over the last decade. In the year 2000, there were only 624 CPMs. By 2009, this number had risen to 1645 (I. Darragh, North American Registry of Midwives Chairperson, written communication, December 2012). In addition, 25% of contributors to the project did not participate continuously throughout the study period, enrolling and disenrolling as they temporarily or permanently exited project participation and/or clinical practice. The system in place between 2004 and 2009 did not allow us to track participation trends closely enough to provide further detail about how many or what types of midwives contributed data throughout the study period versus dropping out early or entering late. The credential of the midwife contributor is also unknown in 4.6% of cases. Although we diligently sought to identify credentials for all contributors who left this portion of the enrollment form blank (through communications with the North American Registry of Midwives, for example), the fact that more than 4% remain unknown limits our ability to describe the project's contributor base. Furthermore, in a recent survey of currently credentialed US CPMs, 13.8% of participants reported having attended no births as a primary midwife in the previous 3 years, despite maintaining a current credential (M. Cheyney, PhD, CPM, LDM et al, unpublished data, February 2013). A final limitation stems from the regulatory environments in some states that restrict midwifery practice, particularly for CPMs. In states without a mechanism in place to census such practitioners, the contributor denominator is unknown. Taken together, these factors make it difficult to calculate the captured proportion of all potentially eligible contributors. Based on examination of MANA Stats enrollment records and data from the North American Registry of Midwives on the number of CPMs by year, our best estimate is that between 20% and 30% of active CPMs contributed data between 2004 and 2009. Although not ideal, especially when vital statistics data are ineffective at capturing accurate outcomes, this rate of participation is comparable to other recently reported midwife-led birth benchmarking projects.[24, 25, 44]
The data entered into the MANA Stats system come from medical records, which have some known limitations when used for research.[45, 46] The main limitation is that medical records are kept to facilitate clinical care, as well as for billing and liability purposes, without thought to future research questions that may be asked of the data. If a given condition is not reported in the medical record, it does not necessarily mean that it was not present, but may be that it was not documented. However, we expect that nearly all of the variables reported here would have been accurately documented in client records because of their importance to clinical care. In addition, in most cases the midwives themselves, not a third-party records abstractor, entered the data, increasing accuracy. Furthermore, for mode of birth, intrapartum transfer to the hospital, etc, our prereview/postreview analysis showed no significant differences, suggesting that the key variables were initially entered into MANA Stats from midwives’ medical records with a high degree of accuracy.
When the 2.0 research dataset was closed in 2011, it included some incomplete forms (2456 incomplete records out of a total of 27,304 logged, or 9.0%). Throughout the project, midwives received automated e-mails every 6 weeks reminding them to complete their forms. In addition, one year prior to closing the 2.0 dataset, all midwives with incomplete entries were contacted by the director of data collection for the DOR and encouraged to complete their records with the assistance of a data doula. While many did, a small number of midwives expressed the desire to leave the project rather than to complete data entry. It is possible that some selection bias was introduced by dropping all incomplete forms from the dataset, although our analyses of these forms suggest a pattern in which midwives simply stopped participating at some point after falling behind rather than a pattern of episodically uncompleted forms. There is no evidence that providers with incomplete records failed to complete those with adverse outcomes.
Finally, we have no way of assessing whether midwives who participated in this project logged every client who consented to participate. It is theoretically possible that some midwives intentionally excluded some courses of care from their reporting; however, the protocol and software required that clients be logged early in the pregnancy, and with a low-risk home birth population it would be difficult, if not impossible, for a midwife to predict ahead of time which births not to log.
Going forward, the quality of the data would be greatly improved if national or state regulations required mandatory participation in MANA Stats or AABC's UDS (now called the Perinatal Data Registry) for all midwives attending home and birth center births, as these are the only registries currently capturing data on planned place of birth and midwifery-led, process-of-care in the United States. Such regulations are already established for CPMs licensed in 2 states (Oregon, Vermont) and are under consideration in others (Arizona, California, Colorado, Texas, and Washington).