Article type
Year
Abstract
Background: Many screening questions face the problem that, due to low disease prevalence, studies need to have enormous sample sizes in order to detect statistically significant effects on patient-relevant outcomes.
Methods: We report on the results of an IQWiG benefit assessment of neonatal pulse oximetry screening (POS) for detection of critical congenital heart disease (cCHD). We compared two different analytical strategies to assess the benefits of POS: the ‘classical’ intention-to-screen (ITS) analysis using all participants as denominator and the ‘alternative’ analysis using only those participants who were affected by the disease.
Results: Only one concurrent controlled study (de Wahl-Granelli 2009) could be included in the systematic review. Based on all 155,567 newborns (ITS analysis), the study failed to show a statistically significant effect of screening on severe acidosis (OR 0.490 (0.217 to 1.109) P value 0.086). Using only babies with cCHD as denominator (n = 160), the study reported a significant effect on severe acidosis (OR: 0.268 (0.110 to 0.654) P value 0.003).
Discussion: In most screening trials, it is useful to analyze results based on the ITS approach, because the screening interventions themselves affect the prevalence of disease (e.g. by detecting clinically insignificant cases). Thus, using only participants with the disease as denominator might introduce bias into the analysis. In the context of cCHD, however, all affected newborns are destined to die when left untreated, so bias is unlikely. Nevertheless, for comprehensive assessment of a screening intervention, harms have to be examined within the total target population.
Conclusion: It is obvious that the benefit of screening primarily evolves from the treatment of affected people. Therefore, if bias is unlikely, the demonstrated benefit in diseased people may justify the implementation of the program, even if the effect on the screened population fails to demonstrate significance. Therefore, if possible and reasonable, when evaluating screening interventions an analysis of results using the affected population as denominator seems to answer additional relevant questions.
Methods: We report on the results of an IQWiG benefit assessment of neonatal pulse oximetry screening (POS) for detection of critical congenital heart disease (cCHD). We compared two different analytical strategies to assess the benefits of POS: the ‘classical’ intention-to-screen (ITS) analysis using all participants as denominator and the ‘alternative’ analysis using only those participants who were affected by the disease.
Results: Only one concurrent controlled study (de Wahl-Granelli 2009) could be included in the systematic review. Based on all 155,567 newborns (ITS analysis), the study failed to show a statistically significant effect of screening on severe acidosis (OR 0.490 (0.217 to 1.109) P value 0.086). Using only babies with cCHD as denominator (n = 160), the study reported a significant effect on severe acidosis (OR: 0.268 (0.110 to 0.654) P value 0.003).
Discussion: In most screening trials, it is useful to analyze results based on the ITS approach, because the screening interventions themselves affect the prevalence of disease (e.g. by detecting clinically insignificant cases). Thus, using only participants with the disease as denominator might introduce bias into the analysis. In the context of cCHD, however, all affected newborns are destined to die when left untreated, so bias is unlikely. Nevertheless, for comprehensive assessment of a screening intervention, harms have to be examined within the total target population.
Conclusion: It is obvious that the benefit of screening primarily evolves from the treatment of affected people. Therefore, if bias is unlikely, the demonstrated benefit in diseased people may justify the implementation of the program, even if the effect on the screened population fails to demonstrate significance. Therefore, if possible and reasonable, when evaluating screening interventions an analysis of results using the affected population as denominator seems to answer additional relevant questions.