Impact of Postnatal Systemic Corticosteroids on Mortality and Cerebral Palsy in Preterm Infants: Effect Modification by Risk for Chronic Lung Disease

METHODS

The literature was searched for all RCTs of postnatal corticosteroids, and additional unpublished data were sought from authors of all known RCTs.^2–4 The Medline search, updated through May 2004, used the terms “adrenal cortex hormones” or “dexamethasone” or “betamethasone” or “hydrocortisone” or “steroids” or “corticosteroids”; limits “randomized controlled trials,” “human,” and “all infant: birth to 23 months.” Studies of only inhaled steroids were excluded. All RCTs of postnatal corticosteroid treatment versus no treatment for prevention or treatment of CLD in preterm infants that reported mortality and the long-term outcome of CP were included. When there was >1 report of long-term outcomes from 1 RCT, we used data from the report with outcomes described at the latest age of the study subjects. We sought data on the rate of CLD in the control group (defined as requiring oxygen therapy at 36 weeks’ postmenstrual age), the contamination rate (percentage of control group ultimately given corticosteroids), the dose of corticosteroids according to the protocol (in mg/kg dexamethasone equivalent), the age of the infants when corticosteroids were started (first week = early; after first week = later), the age of children at follow-up (in months), whether they were assessed by experts who were blinded to treatment group allocation, and the follow-up rate of survivors (%). Descriptions in the RCTs of the types of participants, interventions, short-term outcome measures, study design, and validity assessment of the studies have already been reported.^2–4

Data were analyzed in RevMan 4.2.2, with the endpoints being mortality before follow-up; CP; and, to allow for the competing risks of death and long-term morbidity, the combination of CP or death. The measures of treatment effect for individual trials included relative risk (RR); risk difference (RD); and, for the meta-analyses the typical RR and typical RD, all with their 95% confidence intervals (CIs). A fixed-effects model was assumed for meta-analysis, with significant statistical heterogeneity defined as P < .05. In addition to examining the effect of corticosteroids on death and CP in all of the trials taken together, we conducted subgroup analyses that were based on the postnatal age of starting corticosteroids (early and later) and on the contamination rate (no contamination, >0% to <35% contamination, and ≥35% contamination).

We examined the relationship between the corticosteroid effect on the combined outcome, death or CP, and the risk for CLD (indicated by the rate of CLD in the control group) by weighted meta-regression analysis in Stata (version 7.0)¹⁵ to test the hypothesis that the treatment effect on survival free of CP would be less favorable in trials that were conducted in populations at low risk for CLD and more favorable in trials that were conducted in populations at high risk for CLD. Meta-regression fits models with 2 additive components of variance, one representing the variance within studies, the other the variance between studies. Variance within studies is related to sample size, and larger studies contribute more to the overall results. We then adjusted for other important study design variables, including timing (early versus later) and total dose (≥3.0 mg/kg, <3.0 mg/kg) of corticosteroids, the contamination rate (<35%, ≥35%), the follow-up rate (≥90%, <90%), assessment by blinded experts (yes, no), and age of follow-up (≥36 months, <36 months). These cutpoints were chosen either to be around the median (dose, contamination rate, and age of follow-up) or because they represented desirable features of follow-up studies (follow-up rate and outcome assessment). We then repeated the meta-regression analyses to examine the relationship between the rate of CLD in the control group and the corticosteroid effect on death and CP separately, without adjusting for the other study variables. For all analyses, P < 0.05 was considered statistically significant.

RESULTS

Meta-Regression Analysis: Relation of Treatment Effect to Control Risk for CLD

Fourteen studies reported the rate of CLD at 36 weeks’ postmenstrual age in the control group; this ranged from 9.5% to 73.8% (Table 1). Meta-analysis of these 14 studies found no significant effect on the combined rate of death or CP among children who were randomized (typical RD: 0.04; 95% CI: −0.02 to 0.09). However, there was a significant negative relationship between the RD for death or CP and the rate of CLD in the control group in these studies (Fig 1): for every 10% that the rate of CLD increased in the control group, the RD for death or CP fell by 3.8% (95% CI: 1.4% to 6.2%; P = .002). The regression line crossed 0 (no effect on death or CP) at a rate of CLD of ∼50%. At control rates of CLD below ∼35%, the lower bound of the 95% CI for the regression line was >0, indicating a significantly increased rate of death or CP with corticosteroid treatment. At control rates of CLD above ∼65%, the upper bound of the 95% CI for the regression line was <0, indicating a significant treatment benefit for this outcome.

• Download figure
• Open in new tab
• Download powerpoint

Fig 1.

RD (%) for death or CP among all participants versus rate of CLD (%) in the control group. Each study is shown by a circle whose area is proportional to that study’s weight. The control rates for CLD in the individual studies are given in Table 1, column 5. Regression line and its 95% CI are shown. Regression equation: Y = 18.7 − 0.38X; P = .002. CS, corticosteroids.

None of the potential confounders (time of starting treatment, dose, contamination, follow-up rate, age at follow-up, and assessment by experts), either individually or in combination, significantly affected the relationship between the corticosteroid effect on death or CP and the control risk for CLD. In an adjusted weighted meta-regression based on 13 studies for which the required data were available, after simultaneously adjusting for each of the candidate confounding variables, a statistically significant relationship persisted between the corticosteroid treatment effect on the combined rate of death or CP and the control rate of CLD. Indeed, in the adjusted meta-regression, the absolute size of the regression coefficient was increased, although the CIs were wider: for every 10% that the rate of CLD increased in the control group, the RD for death or CP fell by 9.1% (95% CI: 1.0% to 17.2%; P = .028).

Meta-regression of RD for death alone against the rate of CLD in the control group showed a nonsignificant negative relationship (Y = 3.7 − 0.17X; P = .11). For every 10% increase in the rate of CLD in the control group, the RD for death fell by 1.7% (95% CI: −0.4% to 3.9%). Meta-regression of RD for CP alone against the rate of CLD in the control group showed a significant negative relationship (Y = 14.1 −0.23X; P = .023). For every 10% that the rate of CLD increased in the control group, the RD for CP fell by 2.3% (95% CI: 0.3% to 4.3%)

DISCUSSION

Corticosteroid therapy was initially introduced in chronically ventilator-dependent infants with the hope of saving lives and possibly improving long-term neurodevelopmental outcome if the severity and duration of CLD could be reduced. As early reports of short-term benefits appeared, there was an increasing tendency to start corticosteroid treatment on infants who were less critically ill and earlier in their nursery stay. This tendency has been tempered considerably with reports of an increase in CP among survivors. However, it is possible that corticosteroids are now being withheld in some infants who might benefit. There may be a point, defined by the chance of developing CLD, below which the risks of corticosteroid treatment on long-term outcome outweigh the benefits and above which the benefits outweigh the risks.

The 20 studies included in this review were all randomized trials. Although all studies reported effects on some long-term outcomes, none was designed primarily to assess long-term effects of treatment. We attempted to obtain as complete a data set as possible by contacting the investigators and requesting any unpublished data concerning long-term outcomes. In the included studies, children were assessed mostly early in childhood, before the diagnosis of CP can be certain,⁴⁶ not always by experts who were blinded to treatment group, not always using explicit and reproducible criteria for the diagnosis of CP, and sometimes with potentially important (>10%) loss to follow-up.

Corticosteroids that are given after birth to preterm infants with lung disease have proved short-term benefits, including reductions in ventilator and oxygen dependence and in incidence of CLD.^2–4 These short-term benefits might be expected to lead to a reduction in adverse neurodevelopmental outcomes associated with CLD in preterm infants. However, among all trials, we found no evidence of benefit of postnatal corticosteroid treatment on the rate of CP, regardless of whether the competing risk for death before follow-up was taken into account. In fact, the direction of effects was generally that of harm, reaching statistical significance for CP among all randomized and among survivors examined.

The subgroup analyses that were based on contamination rates of control groups in the various trials support the interpretation of a causal relationship, overall, between postnatal corticosteroid treatment and CP, in that this relationship was stronger in the trials with no or less contamination and weaker or nonexistent in the trials with high rates of contamination. In trials with low or no contamination, there was a tendency for corticosteroid treatment to reduce mortality. This pattern suggests that, overall, postnatal corticosteroid treatment, although maybe reducing mortality, causes CP in survivors.

Corticosteroid treatment was associated with a significant increase in CP in the early treatment but not the later treatment subgroup; however, even in the later treatment subgroup, there was no trend toward benefit. Given that preterm infants are more unstable in the first days after birth, when events such as cerebrovascular hemorrhage are prone to occur, it is possible that any adverse effects of corticosteroids might be exacerbated at that time and that the marked adverse effects noted in the early treatment subgroup could be attributed to postnatal age per se.

However, the result of the meta-regression shown in Fig 1 suggests an alternative explanation for the adverse effects observed in the early treatment trials. Postnatal corticosteroids in the doses used might not only have had direct toxic effects on the developing brain^7,8 but also an indirect benefit to the brain by improving lung function. The early treatment trials enrolled infants who were generally at lower risk for CLD. Thus, infants who were treated early might have experienced net harm because they had less to gain in terms of the indirect benefit, but all were exposed to the direct harmful effects of corticosteroid treatment. The significant negative association between treatment effect on death or CP and risk for CLD supports this interpretation. None of the potential confounding variables was significantly related to the corticosteroid effect on the combined outcome of death or CP. Moreover, adjustment for these variables, including the time of starting treatment, did not eliminate the statistically significant negative relationship between the risk for CLD in controls and the effect on the rate of death or CP.

We recognize that there are limitations of our meta-regression analysis to explore effect modification arising from risk for CLD. First, the analysis was limited to a subset of studies (14 of 20) in which the control rate of CLD was reported. Second, in taking the control rate of CLD as a measure of control risk, we used data that were unavailable at trial entry. Future application of these results in identifying suitable candidates for corticosteroid treatment would require accurate prediction of CLD in at-risk populations. Third, we did not have access to individual patient data, which would allow more sensitive analysis of the relations between risk factors known at baseline, subsequent development of CLD, and corticosteroid effect on death or CP. Fourth, the list of potential confounding variables that we examined was limited by the data available and does not include other possible confounders, such as differences between studies in mechanical ventilation strategies used.

The main focus of our study was on CP because this is the outcome that has been most controversial. We have also taken into account the competing outcome of death. Data on the effects of steroids on other neurologic outcomes are also important but are much more limited and are discussed in the existing Cochrane reviews.^2–4 There were too few studies that reported both the effects of corticosteroids on neurologic outcomes other than CP and the rate of CLD in the control group to allow us to examine those associations in meta-regression analyses.

Future research should be directed toward trying to prevent CLD and hence the need for corticosteroids, to more accurate early prediction of CLD⁴⁷ to allow targeting of any postnatal corticosteroids to those most likely to experience long-term benefit, and to determination of the lowest dose of corticosteroid that is effective in reducing ventilator and oxygen dependence and the incidence and severity of CLD, without causing long-term harm. The analyses reported here support future randomized trials of postnatal corticosteroids specifically in preterm infants who are accurately predicted to be at high risk for CLD, with effects on neurodevelopment and mortality as primary outcomes.