Statistics from Altmetric.com
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.
Commentary on the paper by Sundrum et al (see 15)
Social inequalities have a pervasive influence in almost all aspects of health and health service use. In many instances, as in smoking and lung cancer, the inequalities are a confounding variable and strategies of preventive action can address the primary cause. In those conditions where aetiology is uncertain, investigation of social inequalities may point to causal hypotheses and lead to successful preventive action.
Until 2001, the measures of social status used were the Registrar General’s Social Class (RGSC) based on levels of occupational skill or the more detailed socioeonomic status (SES) classification that brought together occupations of similar social and economic status.1 Alternatively, proxy measures for affluence and deprivation based on census enumeration districts were applied to examine the association between incidence and prevalence rates of a large number of diseases. In the majority of instances, the association is an inverse one; the higher the social position or degree of affluence, the lower is the risk of disease.
In population studies of cerebral palsy there has been no consistent association observed. In this issue of Archives, Sundrum and colleagues examine the relation between cerebral palsy and socioeconomic status.2 The authors draw attention to this in the summary of these studies shown in table 1 of their paper. However, in this report of a population based study of cerebral palsy in West Sussex, they position themselves clearly on that side of the fence favouring a strong inverse association between RGSC and cerebral palsy prevalence. The title of the report is a misnomer because only RGSC and enumeration district deprivation quintiles and not SES are examined.
There are two main areas of this report where the observations made and conclusions drawn need to be tempered with caution. Primarily, the validity of the observation that there is a highly significant linear trend in association of RGSC and cerebral palsy needs to be examined. Secondly, if validity is established, the claim that 51% of cases of cerebral palsy were “attributable” to inequality in RGSC must be clarified.
The basis for the conclusion that the odds ratio for cerebral palsy prevalence has a linear association with RGSC arises from the data in table 2 of their paper. Four striking features are evident in this table:
The very low prevalence of cerebral palsy in RGSC 1
The high prevalence of cerebral palsy in RGSC Unclassified
The high prevalence of cerebral palsy in those whose RGSC is missing
The lack of a discernable trend in odds ratio between RGSC 2 through to RGSC 5.
The RGSC 1 and RGSC Unclassified are at opposite ends of the RGSC classification and, in doing the χ2 test for linear trend, the assumption is made that those whose RGSC is unclassified are at the lowest end of the social scale—that is, they are below RGSC 5. This is not a valid assumption and there are many reasons why the occupation may not be provided at the time of registration. It is pertinent that, when the RGSC unclassified group is omitted from the trend analysis, the trend becomes only marginally statistically significant with p = 0.04.
The prevalence of cerebral palsy in RGSC missing is 10.3 per 1000. This is a 4–5-fold increase when compared with general population cerebral palsy prevalence. The potential bias this group may induce becomes even more worrying in the light of the observation that RGSC was missing in 18.8% of cases of cerebral palsy but only in 4.1% of children without cerebral palsy. This difference is hugely significant (p < 0.0001). The RGSC was missing because the families, though resident in West Sussex, were born out of the county and the birth registration details were not available. Furthermore, this group had a disproportionate number of high risk infants (low birth weight, preterm, and multiple births). It is feasible that, if there is concern over the progress of a pregnancy, RGSC 1 mothers may be more likely than others to seek care or be referred outside the county, say to a teaching hospital. If so, there will be a deficit in cerebral palsy cases in RGSC 1 and the bias engendered would negate any social class trend observed.
In table 2, there is no discernible trend in odds ratios between RGSC 2 and RGSC 5. A χ2 for linear trend limited to these RGSC categories is non-significant (p = 0.94). It is clear that including RGSC 1 and RGSC Unclassified at opposite ends of the social spectrum is sufficient to convert a statistically non-significant trend into a significant one. However, the assumption that RGSC Unclassified is at the lowest end of the social spectrum is not valid and the potential for bias, particularly pertaining to RGSC 1, arising from the failure to record the occupation in a high proportion of children with cerebral palsy, means that any social class trend in the data needs greater justification.
As there is no trend within RGSC 2–5, it would be appropriate to test the null hypothesis that there is no difference in cerebral palsy prevalence of 1.1 per 1000 in RGSC 1 and 2.3 per 1000 in RGSC 2–5 combined. The null hypothesis must be rejected because the difference is highly statistically significant (p = 0.002). Is this an acceptable conclusion that RGSC 1 is somehow protected from cerebral palsy? If so, it is a unique observation and needs justification.
Cerebral palsy is not a single nosological entity. It is recognised that there is an inverse relation between birth weight and cerebral palsy prevalence due, in part, to the risk of very preterm infants sustaining perinatal hypoxic-ischaemic cerebral impairment. Also, the risk of preterm delivery is increased in the lower social classes. Therefore, in this preterm group, it is consistent that a social class gradient in cerebral palsy is observed. Indeed, after adjustment was made for birth weight and gestational age, the social class gradient in odds ratios was no longer significant.
Another group of cerebral palsies where a social class gradient may be expected are those acquired postneonatally, usually as a result of accidental and non-accidental head injury and severe neurological and other infections. These comprise 3–13% of all cases.3 The authors were unable to differentiate these postneonatally acquired cases in their database but, if they could be excluded, any social class trend in the remaining data would be limited further.
Finally, the claim is made that 51% of cases of cerebral palsy were statistically “attributable” to inequalities in RGSC. This calculation is based on the assumption that a cerebral palsy prevalence of 1.1 per 1000 in RGSC 1 is a valid estimate and doubts over this estimate already have been referred to. The calculation also assumes that the prevalence of cerebral palsy in all other RGSCs could be reduced to 1.1 per 1000. In addition to the validity of these assumptions being suspect, although the authors make it clear that the attribution is statistical, the reader frequently considers that the association demonstrated infers causation. This is not so; any observed epidemiological association cannot be assumed to be causal. Hill enumerated several criteria that justified an epidemiological association being considered causal, one of which is that the association be consistently observed in a variety of studies in different locations.4 Such consistency is lacking in studies reporting the association of cerebral palsy with social inequalities.
Whether or not socioeconomic status has a role in cerebral palsy is contentious and the authors’ attempt to shed light on the problem is admirable, particularly since they were endeavouring to use routine data sources to address the issue. Efforts to replicate the study in other districts based on similar data sources are highly recommended.
Commentary on the paper by Sundrum et al (see 15)