Jump to comment:

• Response to letter re: ‘Combinations of subgroup analyses and impact on results’
  Ulrich H. Thome
  Published on: 29 May 2018
• Combinations of subgroup analyses and impact on results
  Eden A Andrew, James Holberton, Gillian Opie and Andrew Watkins
  Published on: 29 May 2018

• Published on: 29 May 2018

  Response to letter re: ‘Combinations of subgroup analyses and impact on results’

  • Ulrich H. Thome, Prof. Dr. med. Division of Neonatology, Department of Women's and Children's Medicine, University Hospital of Leipzig

  We thank the letter authors for commending most of our protocol decisions. A multicenter trial is always associated with a number of compromises, e.g.  between standardization and freedom of therapy, between insufficient and overzealous data collection, and between too few and too many exploratory statistical tests.

   

  For detecting BPD, we used criteria that included all cases with requirement of supplemental oxygen or mechanical support at a postmenstrual age of 36 weeks. This definition was the same as moderate or severe BPD in the more recently formulated consensus definition, and has been used in many other previous trials, testing ventilation modes, high-frequency ventilation, steroid use, permissive hypercapnia, and many others. This made our results comparable to previously published data.

  ...Show More

  We thank the letter authors for commending most of our protocol decisions. A multicenter trial is always associated with a number of compromises, e.g.  between standardization and freedom of therapy, between insufficient and overzealous data collection, and between too few and too many exploratory statistical tests.

   

  For detecting BPD, we used criteria that included all cases with requirement of supplemental oxygen or mechanical support at a postmenstrual age of 36 weeks. This definition was the same as moderate or severe BPD in the more recently formulated consensus definition, and has been used in many other previous trials, testing ventilation modes, high-frequency ventilation, steroid use, permissive hypercapnia, and many others. This made our results comparable to previously published data.

   

  Few authors have previously chosen to limit their BPD definition to the severe cases only. The letter's authors are correct that when analyzing the severe BPD cases separately, there were significantly more cases in the high PCO2 target group in comparison to the control group, which may indicate harm associated with the high target group, especially as BPD is also associated with other adverse outcomes. However, the numbers were small and a possibility of just seeing random noise remains. Specifically, the cited difference of more cases of severe BPD in the high target group was in part compensated by fewer infants in the high PCO2 target group having moderate BPD. If the high PCO2 target was consistently harmful, there should have been both, more infants with severe BPD and more infants with moderate BPD, in the high target group. This situation, with opposite differences regarding the moderately and severely affected infants, increases suspicions that the significant difference regarding severe BPD may be just random noise, and strongly cautions against over-interpreting this piece of data. The risk of random noise being incidentally found statistically significant increases with the number of exploratory analyses performed.

   

  Following the suggestion in the letter, we have also recalculated the logistic regression analysis with 'severe BPD' replacing 'moderate or severe BPD'. The results were similar to the ones published: a significant association of 'severe BPD' with PDI<70 (p=0.13) or MDI<70 (p=0.12) was not found. Logistic regression analyses were not used to estimate the magnitude of risks, but rather, to identify important risk factors. For this purpose, logistic regression analyses are well-suited.

   

  The letter's authors further point out, that permissive hypercapnia is increasingly used as a ventilation strategy. We hope, that this is not the case, since our trial certainly does not provide a scientific basis for such a management change. While there is some evidence that mild hypercapnia may be associated with minor benefits, higher PCO2 targets, as the letter authors point out correctly, have not been shown to improve outcome. This holds also true for our trial, where all trends pointed in the opposite direction, and a significant increase of NEC was found (see Thome UH et al., Lancet Respir Med 3: 534-43, 2015). We have also carefully tested multiple subgroups in search for conditions in which the high PCO2 target may be associated with significant outcome differences in one direction or the other. We did not find a singlesubgroup that showed benefits associated with the high PCO2 target. The most striking result in this search was found in the subgroup of infants with the most severe lung disease. These infants had a significantly worse outcome when randomized to the high PCO2 target, a finding that had been included in our initial publication (see also Thome UH et al., Lancet Respir Med 3: 534-43, 2015).

   

  Since significant differences were only found in subgroups or when using non-predefined outcomes, harmful effects of high PCO2 targets, as used in our trial, cannot be proven, although we sympathize with the view that some results may suggest this interpretation.

   

  Prof. Dr. med. Ulrich H. Thome

  Facharzt für Kinder- und Jugendmedizin / Neonatologie / Pädiatrische Intensivmedizin Leiter der Selbständigen Abteilung Neonatologie Universitätsklinik für Kinder- und Jugendmedizin Zentrum für Frauen- und Kindermedizin Liebigstraße 20a, Haus 6

  04103 Leipzig

  e-mail: ulrich.thome@medizin.uni-leipzig.de

  Show Less

  Conflict of Interest:
  None declared.

  • Back to top
• Published on: 29 May 2018

  Combinations of subgroup analyses and impact on results

  • Eden A Andrew, Medical practitioner Mercy Hospital for Women
  • Other Contributors:
    • James Holberton, Neonatologist
    • Gillian Opie, Neonatologist
    • Andrew Watkins, Neonatologist

  We read with interest the follow up study by Thome and colleagues assessing neurodevelopmental outcomes of the extremely low birth weight (ELBW) infants from the Permissive Hypercapnia in Extremely Low Birthweight Infants (PHELBI) trial1.

  This study makes an important contribution to the evidence-base on the strategy of permissive hypercapnia for ELBW infants. It is a well-powered, multicentre trial and we commend the authors for the ambitious decision to include only intubated ELBW infants and also the use of a clinician-guided treatment protocol. While the methodology allows some systematic bias, there is strong external validity with a patient population representative of ‘real-life’ clinical practice.

  We question the choice to combine the subgroups with moderate and severe bronchopulmonary dysplasia (BPD) for statistical analysis. In Table 2, we note the non-significant p-value for the combined outcome of moderate/severe BPD of 0.30 and no reported p-values for the individual subgroups moderate BPD and severe BPD. Using the raw data provided in Table 2, we calculate a p-value for severe BPD as significant at 0.01, suggesting an increase.

  There is considerable clinical difference between patients with moderate BPD (requiring FiO2 <30% at 36 weeks or discharge) and those with severe BPD (requiring FiO2 ≥30% and/or positive pressure ventilation)2. Other than increased risk of mortality and respiratory disease, severity of BPD correlates with incr...

  Show More

  We read with interest the follow up study by Thome and colleagues assessing neurodevelopmental outcomes of the extremely low birth weight (ELBW) infants from the Permissive Hypercapnia in Extremely Low Birthweight Infants (PHELBI) trial1.

  This study makes an important contribution to the evidence-base on the strategy of permissive hypercapnia for ELBW infants. It is a well-powered, multicentre trial and we commend the authors for the ambitious decision to include only intubated ELBW infants and also the use of a clinician-guided treatment protocol. While the methodology allows some systematic bias, there is strong external validity with a patient population representative of ‘real-life’ clinical practice.

  We question the choice to combine the subgroups with moderate and severe bronchopulmonary dysplasia (BPD) for statistical analysis. In Table 2, we note the non-significant p-value for the combined outcome of moderate/severe BPD of 0.30 and no reported p-values for the individual subgroups moderate BPD and severe BPD. Using the raw data provided in Table 2, we calculate a p-value for severe BPD as significant at 0.01, suggesting an increase.

  There is considerable clinical difference between patients with moderate BPD (requiring FiO2 <30% at 36 weeks or discharge) and those with severe BPD (requiring FiO2 ≥30% and/or positive pressure ventilation)2. Other than increased risk of mortality and respiratory disease, severity of BPD correlates with increased risk and severity of neurodevelopmental impairment (NDI)3. Analysis of all three subgroups of BPD severity might alter the findings and influence the conclusions drawn.

  Conflicting results were yielded from the regression analyses, which may reflect the choice to cluster moderate and severe BPD together. The authors identified moderate/severe BPD as a risk factor for milder NDI, but did not find a significant association with severe NDI. Further, we question the choice to quantify risk using logistic regression to generate odds ratios, as opposed to other statistical models to calculate relative risk. This strategy has been shown to overestimate the measure of association in RCTs and lead to misinterpretation of results4.

  Permissive hypercapnia is increasingly used as a ventilation strategy. The limited evidence to date does not, however, suggest that this improves outcomes for ELBW infants and there have been trends toward worse neurodevelopmental outcomes. While it is commendable to publish a ‘no difference’ result, we would be interested to know if the data were analysed stratifying for all three subgroups whether the findings might suggest permissive hypercapnia in EBLW confers higher risk of severe BPD and thereby poorer long-term outcomes.

  Dr Eden C Andrew MBBS
  Dr James Holberton MBBS, FRACP
  Dr Gillian Opie MBBS, FRACP
  Dr Andrew Watkins MBBS, FRACP

  1. Thome UH, Genzel-Boroviczeny O, Bohnhorst B et al. Neurodevelopmental outcomes of extremely low birthweight infants randomised to different PCO2 targets: the PHELBI follow-up study. Arch Dis Child Fetal Neonatal Ed. 2017;102(5):F376-82
  2. Ehrenkranz RA, Walsh MC, Vohr BR et al. Validation of the National Institutes of Health Consensus Definition of Bronchopulmonary Dysplasia. Pediatrics. 2005;116:1353-60.
  3. Walsh MC, Morris BH, Wrage LA et al. Extremely Low Birthweight Neonates with Protracted Ventilation: Mortality and 18-Month Neurodevelopmental Outcomes. J Ped. 2005;146(6):798-804.
  4. Knol MJ, Duijnhoven RG, Grobbee DE et al. Potential Misinterpretation of Treatment Effects Due to Use of Odds Ratios and Logistic Regression in Randomized Controlled Trials. PLoS One. 2011;6(6):e21248.

  Show Less

  Conflict of Interest:
  None declared.

  • Back to top