We found that reports of RCTs poorly documented key demographic characteristics relating to SES. Almost all trials reported age and gender of participants. Ethnicity and education level were the next most commonly reported variables. Very few trials reported occupational group, employment status, income or area based measures of disadvantage.
The relevance of any one or more demographic characteristics to the internal and external validity of any individual trial will vary depending on the intervention being trialled. Nevertheless, while CONSORT is not prescriptive as to which baseline demographic variables should be reported, the intent of the statement is to "facilitate clarity, completeness, and transparency of reporting" noting that "explicit descriptions, not ambiguity or omission, best serve the interests of all readers". It seems unlikely that the proportions of papers found to be reporting detailed baseline characteristics in this study are consistent with this stated intent. Clinicians, particularly primary care clinicians, seeking to determine the applicability of trial results to their patients may not be provided with sufficient data to do so.
Of particular note is our finding of the relative lack of reporting of SES of trial participants. This makes it impossible to assess whether low SES groups are adequately represented in trial populations. As noted earlier, there is a range of ways that SES may be important to assessing a trials external validity. It has been suggested that lower SES groups may be under-represented in trials [8, 9] but without adequate reporting of this key characteristic this is unclear.
There are good reasons why people from more disadvantaged circumstances may be under-represented in trials. These groups have difficulties in accessing health care so may also be likely to have difficulty with "accessing" trials. For example, complying with private transport or communication technology requirements of study protocols may be a barrier. Lower levels of literacy or speaking a language other than that used in study documents may tend to exclude people from some trials. Co-morbidity, more common in more disadvantaged populations  is also likely to lead to exclusion from trials with strict inclusion criteria. If subjects from lower SES background are systematically under-recruited into trials, the external validity of study findings to these groups may be limited. Higher health-care need in more disadvantaged populations is well documented. Lower SES patients have higher prevalence of disease, more multi-morbidity, more psychological co-morbidity and illness has a greater impact on their lives . In addition to the recognised "inverse care law" , our results identify a potential "inverse evidence law" which implies limitations in generalisability of trial findings for lower SES groups.
An intervention that has an effect in a trial may have little effect in people of low socio-economic status for many reasons, for example, they may receive many drugs already and they may have low compliance with medicine or whatever treatment has been investigated. They may also be so ill because of co-morbid diseases that the effect would not have been reproduced among them, if a trial had been undertaken which included adequate representation from such groups.
The opposite possibility is equally important. If the under-reporting of SES does in fact reflect under-representation of low SES groups, this may also mean that the effect of an intervention could be underestimated for the very group for whom it is likely to offer the greatest benefit. An intervention might be found to have a marginally significant effect in an RCT with an under-representation of lower SES subjects, where trial participants have an overall lower burden of disease. The marginal benefit identified in such an unrepresentative trial could significantly underestimate the potential benefit to a community starting with higher need and higher baseline risk.
In this study we illustrate how SES characteristics are reported. There may be many reasons for study authors not to report this in Table 1 even though they have collected the data. However, the CONSORT statement is designed to assist systematic assessment of generalizability of results. Time-poor clinicians therefore might rely on Table 1 to assess the relevance of the study findings to their patients.
While our analysis is based on a convenience sample of medical journals reporting RCTs, the papers were drawn from leading general medical journals and if anything we would expect this to lead to a bias towards more complete reporting. Future studies should examine the reporting of sampling frames and could focus particularly on trials in areas where socio-economic background is known to have an impact on the outcome.