The development and evaluation of an outcome predictive score for a neonatal intensive care unit in South Africa

Pieper, C. H. (Clarissa Hildegaard) (2003-12)

Thesis (PhD)--Stellenbosch University, 2003.

Thesis

ENGLISH ABSTRACT: Background The care of children is one of the cornerstones of social philosophy. In first world countries most children survive to adulthood. In South Africa the infant mortality rate is much higher than it should be, if compared to the gross capital income per person. The ability to deliver neonatal intensive care (NIC) in South Africa has decreased in the past decade. Therefore it is necessary to choose which babies will receive care. This choice is mainly based on a birth weight (BW) of at least 1000 grams and or a gestational age (GA) of 28 weeks. The only other variable taken into consideration is antenatal care. International scoring systems, like the Clinical Risk Index for Babies (CRIB) score, have been found lacking in accuracy. Aim: The aim of this study was to devise a scoring system which could accurately predict outcome of individual patients before admission to the Neonatal ICU. Patients and methods: Data on the patients enrolled in the CRIB study (1992-1995) were collected retrospectively for the initial cohort (IC). Variables examined were: Maternal risks like age, parity, type of delivery, prolonged rupture of membranes, syphilis and socio-economic status. Neonatal risk factors like BW, GA, gender, ethnic group, ante natal visits, multiple gestations, place of birth, early or late admission to NIC and the one and five minute Apgar counts. Outcome variables examined were mortality, length of hospital stay, duration of ventilation and the development ofbroncho-pulmonary dysplasia. The scoring system was developed with data from the CRIB cohort. A prospective study obtained data for a validation cohort (VC) (1999-2002). Statistical analysis: Descriptive, parametric and non-parametric methods were used. Kaplan&Meier life tables, multivariate analysis and CART analysis were used. Results: The IC consisted of 455 babies with a mean BW of 1198g and mean GA of 30.3 weeks. The VC included 272 babies with a mean BW of 1169g and mean GA of29.8 weeks. The mean maternal income had changed from R892 in the IC to R613 in the VC. These variables were all significantly different. The mortality rate in the IC was 26.1 % and significantly less in the VC of 21.6% (p<0.05). Variables which were the most valuable in predicting outcome were the BW and GA, which were interchangeable. BW had a 63% predictive value for survival. The only outcome variable predictable was survival. BW, antenatal care, gender, place of birth and maternal income were important predictors. Maternal income of zero however nullified all other predictive variables of outcome. In the Cart analysis of the IC the most important predictors were BW > 1037g, maternal income of less than 1206 South African Rand, antenatal care and gender. Survival could be predicted in 94% of cases. In the VC the predictive accuracy was 84% with the CART analysis. The alternative CART analysis was based on place of birth (babies from outlying areas did better), BW «855g) and gender, but did not improve predictability. Discussion Babies admitted to the NICU in this study are chosen by means of non-validated variables. It remains difficult to identify a single prognosticating variable of outcome as the IC was already chosen and the variables are interdependent. Comparable results were obtained in identifying prognosticators when using different statistical methods. The ranking of the variables differed, but the most important variables were similar. Variables currently used to restrict access to the ICU like poor antenatal care and delivery in a peripheral hospital, are no longer justifiable, because babies with these variables did not have a poorer survival rate in this study. A birth weight of more than 855g has the same survival chance as a baby of 1001 grams, which is the current norm for admission. In conclusion, a method by means of the CART analysis was devised that can predict individual survival by 84% or more which is much better than the 63% achieved by using BW.

AFRIKAANSE OPSOMMING: Die versorging van kinders is die hoeksteen van sosiale filosofie. In eerste wêreld lande sal die meeste kinders volwassenheid bereik. In Suid Afrika is die kindersterfde koers baie hoër as verwag, gemeet volgens die per kapita inkomste. Die vermoë om neonatale sorg in Suid Afrika te voorsien het in die afgelope dekade afgeneem. Daarom is dit nodig om die babas te selekteer aan wie sorg verskaf sal word. Die keuse word meestal gemaak op die baba se gestasie (G) (>27 weke) of die geboorte gewig (> 1OOOg)(GG). Al ander veranderlikes wat in ag geneem word is of die moeder antenatale sorg ontvang het aldan nie. Internasionaal gevalideerde seleksiemetodes soos die CRIB score is getoets in Suid Afrika, maar is onakuraat. Doel Die doel van de studie was om 'n seleksiemetode op te stel wat akkuraat die uikomste van individuele babas sou kon voorspel voor die toelating tot die intensiewe sorg eenheid. Pasiente en metodes Data van die pasiente wat gebruik is in die CRIB studie (1992-1996), is retrospektief ontleed as die inisiële kohort (IK). Veranderlikes wat bestudeer is was: Moederlike veranderlikes soos ouderdom, pariteit, tipe van verlossing, verlengde ruptuur van vliese, sifilis en moederlike inkomste Neonatale veranderlikes soos GG, G, geslag, etnisiteit, voorgeboorte besoeke, veelvuldige gestasie, plek van geboorte, vroeë of laat toelatings, en die Apgar telling op een en vyf minute. Uitkomstes wat bestudeer was is mortaliteit, lengte van hospitaal verblyf, lengte van ventilasie en die ontwikkeling van brongo-pulmonale displasie. Die seleksiesisteem is op die gegewens van die inisieële kohort ontwerp en deur middel van 'n prospektiewe validasie kohort (VK) (1999-2002) getoets. Statistiese analise Beskrywende statistieke, parametriese en nie parametriese toetse is gebruik in die analise. Kaplan&Meier lewenstabelle, multivariaat analise en CART analises is gedoen. Resultate Die IK het uit 455 babas bestaan met 'n gemiddelde GG van 1198g en 'n gestasie van 30.3 weke. Die VK het uit 277 babas bestaan met 'n gemiddelde GG van 1169g en 'n G van 29.8 weke. Die gemiddelde inkomste van die moeders was 892 Suid Afrikaanse Rand (R) en R613 respektiewelik. Al drie hierdie veranderlikes was betekenisvol verskillend tussen die twee kohorte. Mortaliteit in die IK was 26% en betekenisvol minder in die in die VK met 21%. Die veranderlikes wat die waardevolste was naamlik GG en G, kan mekaar vervang. Slegs oorlewing kon voorspel word. Veranderlikes wat die meeste gewig gedra het in die meeste analises was GG, voorgeboorte sorg, geslag, plek van geboorte en moederlike inkomste. Geen moederlike inkomste was die enigste faktor van belang in die voorspelling van oorlewing en het die ander faktore tersyde gestel. Die CART analise van die IK het die vorspellers as geboorte gewig van meer a 1037g, sosio- ekonomiese status (>RI206), voorgeboorte sorg en geslag identifiseer. Die metode het oorlewing korrek voorspel in 94% van gevalle. In die VK het die voorspelling oorlewing korrek identifiseer in 84% van gevalle. Die alternatiewe veranderlikes wat geïdentifiseer is in die validasie kohort was die plek van geboorte (babas gebore in die periferie het beter oorlewing), GG (>855g) en geslag maar die alternatiewe CART analise het nie beter gevaar as die oorspronklike me. Bespreking Babas word op onwetenskaplike gronde gekies om intensiewe sorg te ontvang. Dit blyk onmoontlik te wees om een veranderlike te vind waarop die keuse gemaak kan word, aangesien die inisieële kohort reeds gekies was en die veranderlikes onderling afhanklik van mekaar is. Eenderse veranderlikes is deur die verskillende statistiese modelle geïdentifiseer. Die rangordes het verskil, maar die belangrikste veranderlikes was dieselfde. Al die veranderlikes wat tans gebruik om babas te selekteer vir intensiewe sorg is nie van waarde nie. Dit is duidelik bewys deur die CART analise. Veranderlikes soos voorgeboorte sorg en toelatings uit buite hospitale, wat tans in besluitneming gebruik word, penaliseer die babas volgens die ontleding onnodig. GG van meer as 855g skyn aanvaarbaar te wees in plaas van die huidige 1DOOg. 'n Wetenskaplike metode is met behulp van die CART analise ontwikkel wat die vermoë om die oorlewing van babas te voorspel, van 63% na 84% verbeter het.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/49794
This item appears in the following collections: