Article Text

Download PDFPDF

Extremely preterm births: end-of-life decisions in European countries
  1. M S Pignotti1,
  2. R Berni2
  1. 1Neonatal Intensive Care Unit, Anna Meyer Children's Hospital, University of Florence, Florence, Italy
  2. 2Department of Statistics, “G.Parenti” University of Florence, Florence, Italy
  1. Correspondence to Maria Serenella Pignotti, Anna Meyer Children's Hospital, University of Florence, Florence, Italy; m.pignotti{at}meyer.it

Abstract

Objective To explore the influence of end-of-life decisions (EoL-D) on survival and mortality data in the light of differences reported among European countries.

Design We collected the published data of several epidemiological studies: Epicure, Epipage, Epibel and the Norwegian study performed in the UK, France, Belgium and Norway, respectively. The data concerning the Epibel and Norwegian studies are considered for a preliminary analysis, while the data relating to the Epicure and Epipage studies are compared based on the total published data. The statistical analysis was performed through the class of generalised linear models, and more specifically, through log-linear models. The data considered were the number of babies who died in neonatal intensive care units after active withdrawal classified according to the country and gestational age.

Results The probability of death after active withdrawal was significantly higher at 22 and 24 weeks' gestational age compared with week 25, when considering both countries (OR, 2.35 and 1.29, respectively). For the week 2306, the probability of death after active withdrawal was not significant; however, it is relevant when considering the OR (1.31). When considering the country, there was a higher probability of death after active withdrawal in France than in the UK, or alternatively, with the assumed baseline French parameter, in the UK, there was a lower probability of death after active withdrawal, with a significant OR=0.69.

Conclusions The attitude towards EoL-D could in part explain the differences in survival data of extremely preterm infants and should be taken in mind when comparing international survival rates.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

The data on maternal/neonatal outcomes are an indicator of the health of mothers and infants. Their interinstitutional, interregional and international comparisons are used for monitoring the well-being of the population and to point out the areas of major criticism providing information for the management of maternal/neonatal services.1 This comparison requires a detailed validation because the data could be real or the result of one or several artefacts produced during the data collection that need to be cleared before a realistic interpretation is given.1 A number of factors have a direct impact upon the measurement of neonatal mortality rates: the baseline denominator, the differences in terminology and definitions, the recording and delivery policies – often the consequence of different ethical attitudes, inclusion/exclusion of terminations of pregnancy for fetal anomaly, the setting of the study – interhospital, single tertiary centre, interregional studies.1,,3 The MOSAIC study, a prospective collaborative study in 10 European countries, found large differences in mortality of extremely preterm infants, although it involved countries with similar standards of living and healthcare provision.4 We do not actually know the true reason for this difference, although national legislations and practices vary widely across Europe5 6 and medical behaviour could be impacted by local laws and practical guidelines for treatment.7 8 We supposed that the different attitudes in end-of-life decisions (EoL-D) could affect extremely preterm infants' survival data. To explore our hypothesis, we studied the influence of this aspect on newborn' deaths in reported data by means of a statistical analysis.

What is already known on this topic

  • ▶. A lot of deaths in modern NICUs are accompanied by a decision to withhold or withdraw intensive care.

  • ▶. There are a lot of differences in EoL practices within and between countries.

  • ▶. Direct comparison of international neonatal mortality statistics are potentially misleading due to selection bias.

What this study adds

  • ▶. Cross-cultural comparison of survival and morbidity neonatal data must take into account EoL practices.

Materials and methods

We collected the published data of several epidemiological studies: Epicure,9,,11 Epipage,12 13 Epibel14 and Norwegian study15 performed in the UK, France, Belgium and Norway, respectively. These are well-constructed studies whose strength comes from the geographically based population recruitment of the cohort including all births and the large sample size over a short period, which provide a detailed and accurate picture of the birth and survival of extremely preterm infants in the context of current medical practices. We also considered MOSAIC study, which assessed the variability in mortality and short-term outcome after very preterm birth in European countries, but this study was not included in our analysis because of the unavailability of detailed data for infants <25 weeks of gestation. The data concerning Epibel and Norwegian studies were considered as a preliminary analysis, while the data relating to the Epicure and Epipage studies were compared on the basis of the total published data regarding the neonatal intensive care unit (NICU) phase. More specifically, the data according to the common set of available variables are collected by considering the published data9 12 (tables 3 and 1, respectively); the data consist of counts data, that is frequency in a multi-way contingency table. In addition, the definition of active-withdrawal that the authors of both studies used to estimate the number of deaths preceded by active withdrawal of intensive care (IC),9 12 was not considered as influential in this study, due to being a medical choice in the context of clinical practice.

We applied the collected data in the statistical analysis, which is performed through generalised linear models (GLMs).16 17 As previously applied by several authors,18 19 the class of GLMs, statistical models often used in biomedical research, allows us to better analyse the connections between a natural, social and economic phenomena with respect to some explanatory (independent) variables, identified through factors and variables. In addition, GLMs improve some issues and theoretical problems of standard statistical models by highlighting specific features, such as over-dispersion and distributional assumptions. The data are counts data, that is the number of babies who died in NICUs after active withdrawal and classified according to the country and gestational age; therefore, these counts data were supposed to be distributed as a Poisson distribution and the applied statistical model was the log-linear model, which is the most appropriate model for these data, with link function equal to the natural logarithmic function. Thus, the log-linear model evaluates the counts data of the deaths after active withdrawal as a dependent variable; the two countries at issue and the gestational age are included as independent variables. In general, when aggregated and non-individual data are observed, an offset variable is introduced in order to normalise the fitted cell means. In our case study, the logarithm of “admitted to NICU” is specified as the offset variable. It must be noted that the insertion of an offset variable allowed the data to be normalised since the total number of babies, and not individual data, is observed.

Furthermore, analysis of variance (ANOVA) was performed through the likelihood ratio statistics. ANOVA evaluates the main effect of each independent variable, in this case, the country and the gestational age, on the variability of the dependent variable. The choice of the likelihood ratio statistics (type III) allows us to test the significance of each main effect added in the ANOVA model in terms of scaled deviance. The common variables, relating to the Epipage and Epicure studies, were the following: gestational age (GA), died in delivery room (DIED-DR), live birth (LIVE-B), admitted to a NICU (ADM-NICU), died in NICU (DIED-NICU), survived to discharge (SURV-NICU), death after active withdrawal (WDR) and death without active withdrawal (no-WDR). However, in our study, WDRs are studied according to two main factors: country and gestational age. The latter variable was considered as a discrete variable. In relation to the two European countries and the corresponding cited studies, the data are available according to the gestational age (in weeks), from 22 to 25.

Results

Table 1 shows the characteristics of each study and the percentages of deaths after an EoL-D.

Table 1

Characteristics of each study, percentages of deaths after an EoL-D and survival rates

The results of the statistical analysis are reported by introducing p-values and probabilities in terms of odds ratios (ORs). The application was performed through SAS (SAS V.9.1, Windows platform), by using the CATMOD and GENMOD procedures. The applied log-linear model fitted very well; we evaluated this statistical model through two criteria for assessing the goodness of fit: the scaled deviance and the Pearson's χ2, which were not significant. In addition, we evaluated the problem of the presence of overdispersion by also considering a log-linear model with a non-fixed dispersion parameter. In the latter, the fit was not improved with respect to the chosen log-linear model with a dispersion parameter equal to 1. Moreover, when considering the negative binomial distribution as a mixture of a Poisson distribution and a Gamma distribution, in order to take the overdispersion into account, the variance of the dependent variable changes and includes the dispersion parameter ϕ. The estimates of coefficients are generally different, but “it may be shown that the two sets of parameter estimates, one based on the negative binomial likelihood and the other on the Poisson likelihood, differ by a term that is Op−2) for large ϕ”, as reported.16 It must be noted that, in our case study, we do not have a large ϕ and a well-fitted model is estimated with ϕ=1. We found that the medical attitude towards EoL-D is significant with p=0.01 in the UK with respect to the French attitude: the probability, in terms of OR, for an UK baby to die after active withdrawal is 0.69 compared to a French baby treated for the same purpose. Conversely, for a French baby, the probability to die is higher compared to the probability of a UK baby (OR=1.45). Therefore, a significant difference exists between these two European countries in the EoL-D when the parameter estimates are considered: in the UK case, the negative estimate of –0.37 expresses a lower probability (OR=0.69) of death in the NICU after active withdrawal in comparison to France. This result is confirmed by applying the ANOVA, where the country turns out to be a significant factor (p=0.01). In both countries, when considering the gestational age, included as a discrete variable in the log-linear model, the results are significantly different when the four classes are evaluated. In comparison with week 25, the parameter estimate (0.85) of 22 weeks' gestational age is highly significant (p=0.001); therefore, an infant with a gestational age of 22 weeks has a high probability, two times higher (OR=2.35), of death after active withdrawal compared to an infant with a gestational age of 25 weeks. The OR for the gestational age of 23 weeks is 1.31 with a non-significant p value, while we obtain a significant p value (0.05) for the gestational age of 24 weeks, with OR equal to 1.29. In the ANOVA application, gestational age presents a highly significant main effect, with p value equal to 0.01.

Discussion

As pointed out by several authors,1,,3 some selection bias could affect infants' survival and morbidity data and lead to striking differences: the denominator (all births, live births, exclusion/inclusion of lethal malformations or late pregnancy termination, admission to neonatal IC, exclusion of outborn infants); reported survival (at discharge, at 12 months, at 2 years corrected age, etc), differences in the population's characteristics (maternal complications, antenatal steroid use, sex, surfactant use, multiple births, race, etc), the accurate identification of all infants as live born in the case of few signs of life, and the setting of the study. The MOSAIC Study found a large variability in the in-hospital mortality rates in 10 European countries, although standardised for gestational age and sex; ranging from 18.3% (Hesse, Germany) and 25.3% (northern region in the UK) to 50.8% in the Polish regions and 57.9% in The Netherlands for infants of 24–27 weeks' gestational age.4 This variation in survival and morbidity rates between units and countries could be partially explained by differences in EoL-D policy as earlier revealed by Chan et al (inter-hospital variation in delivery room survival rates – range: 41–82%)20 and by the Epibel Study (range: 33–75%).14 Moreover, Lorenz et al showed that a near universal initiation of IC in a population-based cohort of extremely preterm infants (23–26 weeks' gestational age), born in central New Jersey, compared with selective initiation of IC for a comparable group of extremely preterm infants in The Netherlands, was associated with a doubling of survival (46% vs 22%) and a four-time higher prevalence of disabling cerebral palsy among survivors (17.2% vs 3.4%).21 Barton and Hodgman22 found that the contribution of withholding or withdrawing care to neonatal mortality in their unit, on a 10-year period, was 72% of total births: 44% due to “care non-initiated”, 28% to “care withdrawn”, while total care until deaths occurred in 28% of cases. Furthermore, he showed that the contribution of EoL-D to neonatal mortality increased markedly over the 10-year period especially in the group of the smaller infants. A similar number was reported by Partridge and Wall: 73% of the deaths in their unit occurred as a result of selective non-treatment or withdrawal of IC.23 On our own experience in Italy, most of the non-surviving babies die while still on full IC (MS Pignotti, R Agostino, I Barberi et al, unpublished data) contrary to what happen in other countries.24 25 Nevertheless, there are many important differences in practice among units and physicians in Italy26 as well as across Europe and the world.7 8 In Italy, a recent document by the Health Council, an advisory body to the Ministry of Health, invites the physicians to resuscitate every infant showing any kind of life signs, with an attempt to sustain life at any cost.27 This behaviour seems to be very unusual when compared with other countries and with Scientific Societies Official Recommendations.7 In the USA, the Born Alive Infant Protection Act may constrain resuscitation options offered to parents.28 Governmental documents such as these could be thought to have an impact on medical behaviour in an attempt to anticipate medicolegal threats.28 29 In the Epipage study, the limitation of IC leads to 81% and 54% of deaths following EoL-D in infants with gestational age of ≤24 weeks and ≥25 weeks, respectively, for the most in the delivery room (89% vs 60% in the NICU at ≤24 weeks; 64% vs 49% at ≥25 weeks).9 In the Epibel study, 39% of the non-surviving infants died after the decision of withdrawn IC.14 In the Epicure study, 55% of the deaths in the NICU were reported in relation to an EoL-D.9,,11 The authors stated that the planned withdrawal of IC may have influenced the pattern of morbidity in survivors, although many of these infants would have been likely to die anyway. In the light of these considerations, we were interested in understanding whether EoL-D in the NICU could affect survival and mortality data and explain some of the differences in mortality between European countries. We used a statistical model to sustain our hypothesis. We took into account four of the main epidemiological studies performed in recent years in European countries on survival and morbidity of preterm births and chose two of these whose full detailed data were available. We found interesting results showing that the probability of death for NICU-WDR was significantly higher at 22 and 24 weeks' gestational age compared with week 25 (p=0.001 and p=0.05, respectively) when considering both countries. For the 2306 weekers, the OR is 1.31, showing that there is a relevant probability of death for NICU-WDR in this class of gestational age with respect to week 25; nevertheless, the non-significant p value (0.1026) could be explained by the smaller number of babies in this group. When considering the country, there was a higher probability of death for NICU-WDR in France than in the UK, or alternatively, with the assumed French baseline parameter, in the UK, there was a lower probability of death for NICU-WDR, with a significant p value (0.01). Thus, in our opinion, the difference in medical behaviour and attitudes, influenced by the different cultural, social, and legal background of each country, could have an impact on survival and morbidity data of babies. As pointed out by the authors of the Epicure study,9 irrespective of the planned withdrawal of IC, many of these infants would have been likely to die anyway, regardless of IC; however, this seems doubtful in light of the work of Lorenz et al21 and the ever-increasing technological resources in the IC of newborns.

Conclusion

EoL-D may occur at different stages in time: before birth, in the delivery room, in the NICU when the treatment appears futile or the neurological prognosis too poor. The published data show strong differences among European countries in survival rates,4 as well as in the frequency of EoL-D.30 Our statistical results, obtained by comparing UK and France, seem to suggest a significant influence of EoL-D on neonatal probability of death during the NICU phase. This impact could be assumed to be even higher when considering other countries with very different customs, political and religious backgrounds and legislation such as northern and southern European countries. In our opinion, the EoL-D could in part explain the differences in survival data of extremely preterm infants among different countries and should be considered as another potential selection bias to take into account for data interpretation.

Acknowledgments

Our thanks to Ms Mary Susan Cadby for her revision of the English language aspects of the paper.

References

Footnotes

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.