Statistics from Altmetric.com
Patients with congenital heart disease (CHD) have been reported by many authors to have high rates of hospitalisation, morbidity, and mortality associated with respiratory syncytial virus (RSV) lower respiratory tract illness.1–3 However, in a recent paper in Archives of Disease in Childhood, Duppenthaler et al reported a substantially lower incidence of RSV hospitalisation in patients with “haemodynamically significant” CHD.4
They suggest that the rate of hospitalisation in their population of patients from the Canton of Bern, Switzerland was as much as four times lower than rates previously reported in the United States. Based on these results they concluded that the unrestricted use of palivizumab to prevent RSV hospitalisation was not justified.
There are several possible methodological reasons for the disparity in RSV hospitalisation rates in the calculations of both the numerator and denominator. With respect to the numerator, Duppenthaler’s methods would miss all of the nosocomial RSV disease. Furthermore, ascertaining the true incidence of RSV hospitalisation would require that all CHD patients admitted to the hospital undergo RSV screening, as was done in the international multicentre trial, not just those with symptoms judged typical of RSV.5
Finally, in a previous paper by the same authors in the first four years of the study (1997/98–2000/01),6 12 of 497 patients studied aged <5 years were identified with CHD compared to 6 of 449 aged <2 years in this study. Since the previous study encompassed children under the age of 5, the difference of six patients between the first study and this one would imply that children who were hospitalised were between the ages of 2 and 5 (making a strong case for palivizumab prophylaxis in that age group), or they were deemed to have haemodynamically insignificant heart disease (making a case for prophylaxis in this group or questioning the definition of haemodynamically significant heart disease).
With respect to the denominator, the author used the International Classification of Diseases (ICD) coding as a screen for patients with “haemodynamically significant” CHD for entry into the cardiology registry. However, the ICD system does not allow for severity adjustment and therefore does not distinguish between haemodynamically significant and insignificant disease. In our recent multicentre trial, we defined haemodynamically significant CHD as patients with cyanotic CHD, single ventricle physiology, or those with acyanotic CHD that required medical therapy.5 Overall, this would account for only about 35–40% of all the CHD patients. This would have significantly reduced the number of patients in the denominator, increasing RSV hospitalisation incidence. Secondly, no attempt was made to verify the accuracy of screening utilizing the ICD system, which might have resulted in missed patients with CHD due to inaccurate and/or inconsistent coding. Third, the calculations of child-years of observation, as written, was assumed to equal the number of child-years. This would be true if every child were born on 1 July; however, if births are spread out over 12 months the number of child-years should be halved.
Thus, if the numerator were increased (for the reasons stated above) by perhaps one and one-half to two times and the denominator were decreased by two times (even just to account for the incorrect child-years of observation), the actual calculated hospitalisation rate per 100 child-years would be between 3½ and 4 times higher than the rates quoted in tables 3 and 4. These rates would then be comparable to the rates of CHD in the USA.3 In support of this statement is a brief recalculation of their data. The total birth cohort observed over six years was 54 947. There were 813 RSV hospitalisations which would give a rate of 29.6 per 1000 children per year, similar to other developed countries.4 Finally, the calculation of relative risks for hospitalisation uses a referent group that is not low risk, but includes children with prematurity and chronic lung disease, that would unfairly bias the relative risk in a lower direction. The correct referent group would be the low risk group as was done by Boyce and colleagues.3
We would agree with the authors that unrestricted use of palivizumab in CHD patients is not warranted. The intention was never to use the drug indiscriminately in CHD patients as evidenced in the cardiac trial that restricted its use in truly haemodynamically significant young CHD patients.5 We disagree with the use of an NNT analysis to justify this statement. An NNT analysis only factors in the cost from a single RSV hospitalisation. But the CHD infant with RSV is likely to incur additional morbidity and mortality related to future hospitalisations and/or treatment, especially when it comes to surgical correction, and thus raises the cost of care.7 Also, NNT analysis takes only a payer’s perspective, and ignores the societal component of pharmacoeconomics. As healthcare providers, it is our responsibility to use costly drugs in a responsible manner, while also ensuring that these patients receive the treatment/prevention from which they would clearly benefit.
We greatly appreciate the interest of Feltes and Simoes in our study, but we are somewhat surprised by the intensity of their allegations. The large number of unrelated flaws they claim to find in the methods we used suggests that Feltes and Simoes have a fundamentally different view of how these data should be interpreted. In our opinion, their critique is mostly unjustified and requires a firm rebuttal.
It is true that nosocomial infections were not included in our study.1 We did not claim otherwise. However, adding for instance a 14% rate of nosocomial RSV infections as reported for the placebo arm in their study (9 of 63 cases),2 would not translate into a major change in the calculated RSV hospitalisation incidence.
It is also correct that RSV tests were not conducted in CHD patients admitted for reasons other than respiratory tract disease, but it is incorrect to claim that RSV testing was performed only in patients with typical RSV symptoms. Symptoms and signs leading to RSV testing are detailed in the method section of our paper1 and encompass the vast majority of presentations caused by RSV. Furthermore, it is unclear why Feltes and Simoes here compare our observational study with their randomised controlled trial (RCT),2 which obviously led to a more complete case catchment, but may not reflect real life—a well known limitation of RCT. It would have been more appropriate to compare our study with the Tennessee Medicaid Study by Boyce and colleagues,3 which we used as the comparator study. In this study, however, only 6% of cases were specifically coded as RSV infection; 94% were coded as “bronchiolitis”.3 Thus, it appears obvious that our case catchment was not inappropriately insensitive. It is also worth mentioning here that our method of diagnosing RSV had undergone in-house validation.4
In our previous study,5 we did not focus on CHD cases. During the entire six year study period described in the present study, there were 21 hospitalisations among CHD patients <24 months of age, 11 of which were considered by the cardiologist as haemodynamically insignificant. There were an additional three cases in children >24 months of age, all haemodynamically significant. During the first four years of the study period covered also by our previous study,5 there were 12 hospitalisations, all in children <24 months of age, six of which were considered haemodynamically significant. During the subsequent two years, there were 12 cases. Of these, nine cases occurred in children <24 months of age and were haemodynamically significant in four.
We did not—as Feltes and Simoes apparently assume—use ICD codes for case catchment. As stated in the method section,1 the ICD-10 codes Q20–Q26 describe CHD patients included in the database. The large groups of haemodynamically insignificant VSD, ASD, and PDA are not included. All cases included in the registry were “CHD requiring medical therapy”. It follows that the definition of haemodynamically significant CHD used in our study to create the denominator is quite similar to the definition used by them.2 Their claim that the true denominator of CHD in our population was only 35–40% of the figure used, is incorrect in our opinion. Again, our main comparator study3 used a less stringent definition of CHD (ICD-9 codes 745–747) which did not address the issue of haemodynamic significance, and was thus more likely to report estimates of low precision.
Feltes and Simoes also claim that we got it wrong with the “child-years” and thus overestimated that denominator by a factor 2. Objection! We agree that RSV exposure only occurs during approximately half of the child-years, and that total child-years should be divided by factor 2 for calculation of incidence per child-year of RSV season. However, we compared our incidence rates with those of Boyce and colleagues,3 which were already corrected for this discrepancy (that is, our data in table 3 were compared to the column entitled “Hospitalisations per 1000 children” in table 1 of Boyce’s paper, p. 867,3 which were multiplied by factor 2). Had we truly committed the mistake claimed by Feltes and Simoes, we would have compared our data with the preceding column in the said table, which is entitled “Incidence” (meaning: hospitalisations per 1000 child-years of RSV season).
Reduction of our denominator by factor 2 was indeed necessary, when children <6 months of age only were investigated. This, however, we did for calculation of the figures in table 3 and we explicitly stated that we did so in the text on page 963.1 Thus, this allegation again is incorrect in our opinion.
It is true that we used the entire population of non-CHD patients as referent, because we did not have data of sufficient quality for analysis of other individual risk factors. This was clearly stated in the manuscript. However, to claim that such a comparison is “unfair” is difficult to understand, because (1) for comparison we used Boyce’s raw data3 to calculate non-CHD rates in their population, and (2) in Switzerland, palivizumab has been recommended for children with severe BPD only.
Thus, we believe that it does make sense to compare CHD patients to all others who do not receive palivizumab. The very small group of children with severe BPD makes no substantial difference here.
We agree with Feltes and Simoes that NNT should not play a major role when it comes to providing optimal care for children with CHD. This is one reason why the new 2004 Swiss recommendations for the administration of palivizumab include children <1 year of age with surgically uncorrected, haemodynamically significant CHD and cyanotic CHD or severe pulmonary hypertension or cardiac failure,6 as soon as the distributor of palivizumab successfully applies for mandatory coverage by the health insurance companies. If, however, resources are limited, and they increasingly are in many European countries, cost-effectiveness analyses including NNT do play a role when authorities have to weigh different new interventions against each other.
In summary, we believe that Feltes and Simoes create a largely incorrect worst case scenario of what could have gone wrong with our study. As elaborated above, we believe that our data are correct and—with the limitations described in the paper—reflect the current epidemiology in the study area.