Introduction.
Recently, a new classification of social class, the UK National Statistics Socio-economic Classification (NSSC), has been introduced to replace the older Registrar General's social classification (RGSC). It is well known that social class differences may explain some health inequalities. The new classification uses a 7 point classification (table 1) and is explicitly based on differences between employment conditions and relations¹ and assigns social class based on occupational title and responsibilities over other workers. Recent work by our group has explored relationships between a patient's innate desire for drug information (the Intrinsic Desire for Information scale, IDI) with demographic and disease related variables^2,3. During recent analysis of a pooled database of general medical inpatients, anomalies in the coding of social class, using the NSSC, were identified. The aim of this study was therefore to test coding reliability using the NSSC during field work, exploring the possible effects on demographics analysis using the IDI scale.

Method.
A pooled database containing 630 inpatient cases recruited from three London teaching hospitals was used. Each case represented a standardised bedside interview, comprising the 6-item version of the IDI scale^3,4 together with demographic data that included the patient's description of occupation, or immediate past occupation if they were retired. Students and 'housewives' were excluded. The occupation titles were extracted from the database, together with case number and gender. Pairs of coders were asked to code the job title for a random selection of around 80 cases using the NSSC. The pair reached a consensus for each case. Each pair also coded another random set of cases in a similar manner, blind to the first coding. Cases thus received two independent coded variables which described social class according to the NSSC, in addition to the original code assigned from the primary data collection. Variables were compared using Cohen's kappa, a measure of agreement which takes account of the calculated chance agreement between coders.

Results.

With students and 'housewives' excluded, n=474. The kappa value for the dual independent NSSC coding was 0.171 (p<0.001) indicating a low level of inter-coder agreement. To test for anomalies between the coding pairs, kappa values were calculated for the case load of each combination of pairs. These ranged from 0.14 to 0.28 indicating, at best, 28% non-chance agreement. When the NSSC category groups were condensed (into categories 1 and 2; 3 and 4; 5, 6 and 7) kappa improved to 0.27.

Table 1: NSSC categories

Conclusion.
These results show that coders are variable in the assignment codes using patient described job title and the NSSC categories. This has clear implications when conducting health services research.

References.

1. Rose D. & O'Reily, K. 1997. Constructing Classes. ESRC/ONS, Swindon.
2. Duggan C.& Bates I. The development and evaluation of a survey tool to explore patients' perceptions of their prescribed drugs and their need for drug information. Int. J. Pharm. Prac. 2000; 8:42-52.
3. Åström K., Carlsson J., Bates I., et al. Desire for information about drugs: A multi-method study in general medical inpatients. Pharmacy World and Science, 2000; 22: 159-164.
4. Åström K, Carlsson J, Bates I, et al. Information preferences of medical patients: A quantitative and qualitative perspective. Abs: 6th Health Services Research and Pharmacy Practice Conference, The Robert Gordon University, April 2000 (ISBN 1 901085 600)