Methods to Encourage the Use of Antenatal Corticosteroid Therapy for Fetal Maturation:  A Randomized Controlled Trial FREE

A preterm birth occurs in 10% of all live births and is a major cause of neonatal mortality and morbidity.¹ In preterm infants, survival is strongly related to gestational age (GA).^2- 4 Antenatal corticosteroid therapy for fetal maturation is one of the most effective ways to improve the outcomes of preterm births. Randomized controlled trials conducted between 1972 and the early 1990s confirm benefit in most populations.⁵ Antenatal corticosteroid therapy reduces the risk of infant mortality by approximately 30%, of neonatal respiratory distress syndrome by approximately 50%, and of both intracranial hemorrhage and periventricular leukomalacia by approximately 70%. The latter 2 conditions are among the best predictors of long-term neurodevelopmental injury, including cerebral palsy. Net economic benefits are estimated at more than $3000 per treated neonate.⁶

Despite evidence of its effectiveness, use of corticosteroid therapy in eligible infants remained relatively low through the 1990s.⁷ Of particular importance, the therapy was used less often when GA was less than 28 weeks, compared with GA between 28 and 34 weeks.^7- 8 Yet these very low-GA infants were precisely the group that would most likely benefit from antenatal corticosteroids.⁹ Because of these discrepancies, the National Institutes of Health (NIH) with its affiliate, the National Institute of Child Health and Human Development (NICHD), convened a consensus panel in February 1994 to develop practice recommendations for antenatal corticosteroid therapy. The panel concluded that nearly all women who are between 24 and 34 weeks of their pregnancies and are likely to deliver preterm are candidates for the therapy.⁶

In this article, we report results of a randomized controlled trial comparing changes in the use of antenatal corticosteroids associated with publication of practice recommendations (usual dissemination) with those caused by an active, focused dissemination effort (active dissemination), in addition to the practice recommendations. The randomized trial addressed the 3 following questions:

The NIH Consensus Conference reviewed the evidence concerning efficacy of corticosteroids for preterm deliveries of different types.^6,13- 14 The panel recommended that all women at risk of premature delivery, between 24 and 34 weeks' gestation, should receive the therapy, unless there are contraindications because of maternal disease. If PROM is diagnosed, the therapy is recommended for women of up to 32 weeks¹ gestation, unless clinical chorioamnionitis is present. Two corticosteroids appear equally effective: 2 doses of 12-mg betamethasone intramuscularly 24 hours apart, or 4 doses of 6-mg dexamethasone given intramuscularly 12 hours apart. The therapy should be administered unless delivery is imminent.⁶

To understand the reasons for limited use of corticosteroids, we conducted interviews and focus groups with maternal-fetal medicine specialists, neonatologists, and obstetricians around the nation.¹² The panel crafted its recommendations specifically to address the uncertainties and reservations practitioners voiced. For example, the recommendation for practice was fairly simple, because practitioners gave several reasons for delaying the administration of corticosteroids. Also, practitioners overestimated risks and underestimated benefits. The NIH recommendations began with the statement: "The benefits of antenatal administration of corticosteroids to fetuses at risk of preterm delivery vastly outweigh the potential risks."⁶

The design of the randomized trial is shown in Figure 1. To avoid diffusion of the active dissemination treatment to the control group, the unit of randomization and treatment was the hospital. The hospitals were all tertiary care facilities with neonatal intensive care units (NICUs), institutions most likely to adopt these NIH recommendations quickly. Of 30 hospitals invited, 27 participated. Criteria for participation included having at least 100 eligible cases in the baseline year, having no standing protocol, and participating in no other research in antenatal corticosteroids. Albert Einstein College of Medicine (AECOM) of Yeshiva University, New York, NY, recruited 8 affiliated hospitals and all participated. The National Perinatal Information Center (NPIC), Providence, RI, recruited 22 hospitals from its 1994 member network and 19 participated. One NPIC hospital never responded and 1 refused participation. Hospitals were then randomized, but before intervention started another NPIC hospital withdrew, precluding data collection.

The NPIC and AECOM conducted randomization separately for their member hospitals. We assigned hospitals by random number table either to the active dissemination (n=13) or usual dissemination control (n=14) group. National Perinatal Information Center hospitals all were located a substantial distance from each other, so diffusion of the intervention to control institutions was unlikely. Hospitals selected by AECOM were all in the New York area. However, cross-hospital diffusion of the intervention was unlikely because almost no practitioners overlapped across institutions, and residents were shared only by 2 institutions that were both fortuitously randomized to active dissemination. The study was not blinded because physicians in the active dissemination condition were aware of the study, and the leadership of all hospitals (including the chairpersons of obstetrics and gynecology departments) were aware of the condition of assignment.

The NIH Consensus Conference convened in late February 1994. The consensus conference final statement was released on May 27, 1994. The American College of Obstetricians and Gynecologists' opinion statement on antenatal corticosteroids was mailed to its membership in December 1994 and by the end of January 1995, NIH brochures outlining the conference recommendations had been mailed to medical care institutions, universities, medical societies, and obstetricians. In addition, JAMA published the NIH recommendations on February 1, 1995,¹³ and a second publication appeared in a special supplement to the July 1995 issue of the American Journal of Obstetrics and Gynecology on antenatal corticosteroids for fetal maturation.¹⁴ These activities, the available research literature, various lecture formats, and word-of-mouth discussions constituted the way NIH recommendations are usually disseminated and were the way physicians in the control hospitals, the usual dissemination control group, received the information.

We exposed intervention hospitals to an active, focused dissemination effort, tailored to overcome barriers impeding the use of corticosteroids. Active dissemination began in spring 1995. Implementation at the 13 intervention hospitals was staggered from April through mid July 1995 to accommodate supervision of start-up and the scheduling of grand rounds by study investigators. Implementation consisted of 5 components designed to make use of research on adoption of medical practices.^{10- 12,15- 19} The components are as follows:

As is often the case in a decentralized effort, the treatment hospitals varied in the extent to which they implemented active dissemination. Across the 13 intervention hospitals, average attendance at the grand rounds was 63% (range, 33%-93%) of obstetricians managing high-risk deliveries. The proportion of chart reminders inserted in eligible records averaged 73%, and across intervention hospitals, they ranged from 40% to 93%. All institutions included the component of case scenario discussions, and all influential physicians and nurse coordinators responded to feedback on implementation.

For statistical power, our goal was to abstract an average of 162 medical charts per hospital per study year. The charts of 6798 eligible women were abstracted from the 27 institutions, 3516 in the baseline year and 3282 in the after-conference year. We abstracted data on the use of corticosteroids, the timing of their administration, and maternal and infant conditions. To measure hospital practices at baseline, we abstracted maternal medical records from the 12 months immediately before any consensus conference activity or publicity had begun (March 1993-February 1994). Chart abstraction from the 12-month period following the conference began in April 1995 and ended in July 1996. To control for the staggered implementation schedule, record abstraction used the same after-conference time frame for active- and usual-dissemination hospitals in the same geographic regions.

Eligible cases included all women giving birth at 34 weeks' GA or less, including cases of spontaneous preterm labor, PROM, and preterm delivery indicated because of medical conditions. It is important to note that in cases of suspected preterm deliveries, it is not always possible to know which women will deliver preterm. Therefore, some women who did not deliver preterm also received corticosteroids. However, we have no data on these patients, and the sampling frame did not include them.

The number of usable abstracts per hospital averaged 130 in the baseline year (range, 81-168) and 122 in the after-conference year (range, 73-170). We included all eligible cases from AECOM hospitals and sampled 81% of all eligible records from the often-larger NPIC hospitals. Because we wanted to analyze responses to individual cases based on GA, we oversampled the comparatively rare births of neonates between the ages of 24 and 28 weeks' GA. For hospitals with 162 cases or fewer in each year, we abstracted medical records of the entire population of preterm deliveries. For NPIC hospitals with more than 162 cases, we abstracted the entire population of births of neonates whose GA was between 24 and 28 weeks; a random sample of other preterm deliveries was then included, up to 170 cases. Two hospitals had a large census of preterm deliveries; from these hospitals, we selected a simple random sample of neonates of all GAs. Hospital census varied by year, and, therefore, the oversampling had to vary. In the baseline year, 4 treatment and 4 control hospitals included the entire population of low GA cases (31% and 28%); in the after-conference year, 3 treatment hospitals and 5 control hospitals (23% and 36%) included all the low GA cases.

We trained medical data collectors to identify cases in a 2-stage process. To enumerate and sample cases, they first examined logs kept by hospital labor and delivery units. Then, they inspected the records and abstracted those that met criteria. They discarded cases at stage 2 for the following reasons: GA was outside study criteria; GA was not recorded; fetal death was verified on admission; record numbers were inaccurate; records could not be located; and no admission corresponded to delivery information. Based on the log enumeration, we discarded a median of 10.5% of records across hospitals. In 3 hospitals, the proportion of discarded cases exceeded 20%. These 3 hospitals had many cases that appeared eligible on labor and delivery logs, but the medical records listed GA older than 34 weeks or younger than 24 weeks. In these 3 hospitals, between 8% and 13% of the records could not be found.

Corticosteroid administration was influenced both by factors at the patient level, such as maternal diagnosis, and by hospital-level factors, such as the historical rate of corticosteroid use and the hospital-wide dissemination efforts. Analysis addressed both levels because information is lost if either level is ignored. (The physician level of analysis was not available to the study because a condition for hospitals' participation was that individual physicians would not be studied.) The primary analysis was based on a 2-level logistic regression model, with patients as the units of analysis at level 1 and hospitals at level 2.^20- 21 Patient-level variables included the study year (baseline vs after conference); corticosteroids received or not received (not received includes cases in which corticosteroids were ordered but not administered); GA between 24 and 28 weeks (24-28 weeks vs 33-34 weeks); GA between 29 and 32 weeks (29-32 weeks vs 33-34 weeks); and diagnosis of PROM (yes or no). Too few cases of indicated delivery appeared for meaningful separate analysis in the model; thus, the model contrasts spontaneous preterm labor and indicated delivery without PROM against deliveries with PROM. The hospital-level variable is treatment (active vs usual dissemination control). For practical reasons, we estimated a conservative model in which only coefficients for the main effects (study year, PROM, GA between 24-28 weeks and GA between 29-32 weeks) were allowed to vary randomly across hospitals. Nonsignificant main effects were included when they were involved in significant interactions, and only the significant interactions are included.

There were no baseline differences between intervention and control hospitals for the following characteristics: geographic region, median number of active obstetricians, births per hospital, NICU beds, percentage of Medicaid patients, race, PROM diagnosis, GA, and indicated deliveries. Hospital characteristics were generally the same in both the NPIC and AECOM hospitals, so we do not report them separately. A difference between intervention and control cases in the frequency of abnormal fetal conditions or fetal distress was significant at the patient level (χ²=10.73; P<.05) due to the large sample size. It may be a chance effect, given the multiple comparisons between intervention and control institutions.

During the after-conference year, 2 differences in hospital case mix emerged between intervention and control institutions. Based on hospital census (not sampling), treatment hospitals had a larger proportion of the lowest GA cases in the after-conference year (t=3.36; P<.001) because 3 control hospitals reduced their proportion of such cases, while 1 treatment hospital increased its proportion of these cases. In addition, in the after-conference year the average proportion of reported PROM cases increased in all hospitals but increased significantly more in treatment than in control hospitals (t=3.03; P<.006). To examine whether these after-conference differences in PROM and GA would bias the tests of study hypotheses, we conducted 2 analyses. First, we established that there was no significant collinearity between corticosteroid use due to treatment and use due to either PROM or GA, by analysis of the level 2 parameter variation. If these variables had influenced the test of the dissemination treatment, collinearity would have been seen. Second, data were analyzed separately by various levels of GA and whether PROM was present (data not shown). Results are essentially the same as those reported herein. In other words, the change in case mix did not cause increased use of the therapy in the active dissemination group.

In the Table 1, we provide an overview of the percentage of use of corticosteroids by study year and treatment condition at both the patient and hospital levels of analysis. The 2-level logistic regression revealed a significant effect of study year (P<.01). Controlling for other variables in the model, the expected increase in the odds of corticosteroid use after the consensus conference (compared with baseline) is 3.16. In usual dissemination hospitals, use of corticosteroids increased 75% from baseline to after the conference. In the active dissemination hospitals, however, intervention gave a statistically significant "boost" to the effect of the after-conference study year. Active dissemination increased the odds of administering corticosteroids during the after-conference year by a factor of 1.63 (P<.01). In active dissemination hospitals, use of corticosteroids increased 108% from baseline to after the conference, more than one third again as much as in the control hospitals.

Hospitals showed extreme and statistically significant variation in corticosteroid use, which is not explained by the treatment effect. This variation is significant across hospitals in terms of increased use of corticosteroids due to the NIH recommendations (after conference vs baseline) increased use in PROM and increased use at various GAs. Across hospitals, the increase in the rate of corticosteroid use ranged from +5% to +62%. Some hospitals increased corticosteroid use minimally, while others changed their practices radically; some responded to patient characteristics in markedly different ways.

Gestational age and the presence of PROM were associated both with baseline corticosteroid use and with increase in corticosteroid use following the conference. These relationships are shown in Figure 2, which also describes the indicated deliveries separately. In both the baseline and the after-conference years, patients with lower GA were significantly more likely to receive corticosteroids than those with higher GA. Compared with GA between 33 and 34 weeks, the baseline odds ratio was 4.48 for GA between 24 and 28 weeks and 2.95 for GA between 29 and 32 weeks (both P<.01). In addition, the increase in corticosteroid use during the after-conference year was significantly less for patients with GA between 24 and 28 weeks than for those with GA between 33 and 34 weeks (interaction of GA 24-28 weeks and study year, P<.01). Although at baseline use of the therapy was less likely in cases in which PROM was present than when it was not, during after-conference year it increased to a greater degree when PROM was present (PROM by study year interaction, P<.01).

The NIH panel stated that chorioamnionitis and imminent birth are contraindications for the therapy.⁶ In this data set, we could not assess whether chorioamnionitis was diagnosed before or after corticosteroids were administered. However, chorioamnionitis was present in 13.9% of patients receiving therapy in the year before the consensus conference and in 13.4% in the after-conference year. There were no differences between usual- and active-dissemination hospitals in the proportion receiving corticosteroids. "Imminent delivery" was not defined by the panel. However, significantly more deliveries (3.0%) occurred within 2 hours of first corticosteroid use in the after-conference year, compared with 1.8% at baseline (t=2.36; P<.05).

In the 12 months after the consensus conference, use of the therapy increased dramatically among both active- and usual-dissemination hospitals. Few studies of practice recommendations have examined the appropriateness of change.²² Although our conclusions are preliminary, it appears that inappropriate use increased only in the case of imminent delivery, affecting an additional 1.2% of cases in which the therapy was given, a level that is more than counterbalanced by the benefits accruing to neonates. In both time periods, questions about potentially inappropriate use of corticosteroids were present in about 16% of cases, and in most of these, it is likely that their use was appropriate at the time they were given.

The effectiveness of the NIH recommendations is consistent with other research identifying the forces affecting the adoption of practice guidelines, ie, endorsement by the American College of Obstetricians and Gynecologists, a fairly simple recommendation, and absence of alternative or competing therapies.^17,23 By studying tertiary care institutions, the study probably captured the actions of those who would be the first to adopt the NIH recommendations.²³ This is both a strength, in that the study captured the change as it occurred, and a limitation, since less-specialized hospitals were not represented. However, we required a high volume of preterm deliveries for statistical power, which was the deciding factor.

Active dissemination, at a minimum, hastened adoption of the therapy. An additional 170 patients, or 13 per hospital, received corticosteroids in the after-conference year over and above the number of patients who received it at usual-dissemination hospitals. Given the substantial reduction in morbidity and mortality, plus the cost savings per treated neonate, these results are not trivial. Moreover, the effort involved in this educational intervention was relatively low compared with the labor-intensive strategy of one-on-one academic detailing.^10,24- 25

Few studies of practice guidelines have analyzed the costs and benefits of dissemination.²² In actuality, the cost of active dissemination was borne by the study and not by the participating hospitals. However, if the dissemination costs were borne by the hospitals, the costs would vary substantially based on issues such as whether the grand rounds speaker was an in-house expert or one flown in from outside, whether a nurse coordinator could add the intervention tasks to those currently performed, whether additional staff needed to be hired, and whether the influential physician received payment for assuming this role or whether this was simply one of the many functions performed by a department chair or division chief. Most important is whether one assumes additional costs for the practitioner's time to attend educational sessions. Using various assumptions, the incremental cost of active dissemination could vary from $1000 to $16,000 per hospital in the first year. If these costs were spread over the additional 13 treated neonates in the first year, the incremental cost per treated infant would be between $77 and $1231, far less than the $3000 savings per treated neonate, according to the consensus conference.⁶ Since the cost of corticosteroid treatment is low, about $7, we believe active dissemination would substantially reduce costs relating to the care of the neonates.

Findings are mixed concerning the effectiveness of educational strategies. They were probably effective in the present study because it addressed specific barriers and because the overall climate favored a change.^10- 12 As a model for intervention, the active dissemination strategy has certain limitations. As in many dissemination studies, intervention is a complex package; additional research is needed to identify the "active ingredients" to change medical practices. Also, effectiveness has not been determined for hospitals implementing this strategy by themselves, without study oversight.

These findings highlight the importance of a 2-level analysis to preserve information about factors associated with change. Prior studies of practice guidelines generally analyze at the individual level or the organizational level, but not both.^17,19 We did not predict that cases of PROM would see even greater increases in use of the therapy than other maternal diagnoses nor did we predict the high use, even at baseline, among very low-GA infants. These findings will require confirmation in future studies. However, we can speculate about the reasons. First, the study practitioners are located at tertiary care institutions, where it is likely that they have resolved these issues more rapidly than other institutions. Second, delivery was rarely imminent when PROM was diagnosed, but it was more commonly imminent among the women in spontaneous preterm labor. The timing of delivery in PROM cases is therefore under the obstetrician's control to a greater degree, and the therapy could be appropriate for a higher percentage of cases. Finally, the high rate of corticosteroid use in very low-GA infants may be attributable to the progress seen in the outcomes of these infants. In the past, obstetricians may have questioned whether such infants would survive, and therefore did not treat them. With new technology to increase their survival, the decision to treat may be more common, along with greater awareness of the benefits of corticosteroids.

Although the treatment variable explained some of the hospital-level variation, there remain important differences among hospitals in both their initial use of corticosteroids and their responses to the NIH recommendations. Further work is needed to explain why hospitals differed so radically in their response to NIH recommendations. These differences in response to practice guidelines have occurred in other studies, and remain unexplained.¹⁸ Site-level differences are consistent with current understanding of corticosteroid therapy, the use of which varies markedly by institution.^8,26 Not only did the rate of corticosteroid use vary initially, as noted in other studies, but institutions also varied in terms of their response to the recommended change.

The year after the NIH Consensus Conference on Antenatal Corticosteroid Therapy saw a marked change toward increased use of the therapy. An active dissemination effort significantly increased adoption of the NIH recommendations, over and above the effects of the usual consensus conference dissemination methods. We hypothesize that this occurred because conditions were favorable for adoption of the NIH recommendations and because active dissemination was tailored to concerns of obstetricians at the study institutions.