Author's response to reviews Title: Gender, ethnicity, health behaviour & self-rated health in Singapore Authors: Wei-Yen Lim (Lim_Wei_Yen@moh.gov.sg) Stefan Ma (Stefan_Ma@moh.gov.sg) Derrick Heng (Derrick_Heng@moh.gov.sg) Vineta Bhalla (Vineta_Bhalla@moh.gov.sg) Suok Kai Chew (Chew_Suok_Kai@moh.gov.sg) Version: 2 Date: 26 February 2007 see over Author's response to reviews:
26 Feb 2007 Editor BMC Public Health Re: MS: 1103320598113933 - Gender, ethnicity, health behaviour & selfrated health in Singapore Thank you very much for the opportunity to revise our manuscript. We would also like to thank the reviewer for the useful suggestions he has made. We have made revisions to the manuscript taking the suggestions into account, and would like to submit the revised manuscript for re-review and your consideration for publication please. The attached below bears our response to each of the reviewer s suggestions items by item: Compulsory Revisions 1) Self-rated health has been dichotomized so that 76% of the population are compared with the remaining 24%. It would be wise to repeat the analyses using linear regression or logistic regression which allows for at least 3 categories of the dependent variable. Otherwise, the distinction is very crude. Our response: We agree with the reviewer that the dichotomisation is very crude. The two major reasons why we have adopted this dichotomised categorisation of self-rated health as the dependent variable is because we had wanted (1) from a practical viewpoint, to be able to compare our findings with other studies that have also similarly dichotomized self-rated health when exploring risk factors; and (2) from a statistical viewpoint, to obtain more stable estimates from the regression analysis. As suggested, we have performed multinomial logistic regression using self-rated health categorised into 4 categories: very good, good, moderate, and fair/poor as the dependent variable. However, the regression estimates obtained from the regression analyses (data not shown) were very unstable with very wide confidence intervals because of the small numbers in the fair/poor category (79 of 6236 respondents; please see the new Table 1) which led to difficulty in interpreting the findings. Therefore, we would like to maintain the original analysis in the revised manuscript. However, we are open and would leave the decision to the editor and the reviewer whether to include the new multinomial analysis in the manuscript. 2) Page 6 first paragraph: eligible what are the criteria for eligibility? Were there participants who were cognitively unable to participate? Were there proxy interviews?
Our response: The criteria for eligibility used in our study were all Singaporean nationals and permanent residents aged 18 years and above. The sample selection was a 2-stage stratified design. A sample of 11 200 household addresses was selected from the National Database on Dwellings in Singapore, maintained by the Department of Statistics. The primary selection units (PSU) for the first stage consisted of geographic zones while the secondary selection units (SSU) for the second stage comprised household addresses. The PSUs were selected by probability proportional to size, and an equal number of SSUs were taken from each PSU by systematic sampling with a random start during the second stage. The KISH table were then used to identify a respondent (aged 18 years and above) from each selected household. Because this was a community survey and excluded institutionalised individuals (eg those elderly were staying in nursing homes at the time of our survey conducted), we would expect that the elderly participating in our survey would be more likely to rate their health positively (ie towards the very good/good categories). On the other hand, in our study, 33 did not know/were unable to rate their own health, and 1 refused to reply. This represents about 0.5% of the sample (please see the new Table 1). Regarding the question on proxy interviews, we are aware that proxy interviews were used and in many cases, family members assisted the survey subject in responding to the question. (Our database suggests that informants assisted the subject in 30% of respondents.) However, we have no data in our database on the extent to which these proxy interviews were carried out (whether in full, if the informant assisted the subject in responding to the whole survey, or if the informant answered only specific questions that the subject was unable to answer). We apologise for this. We expect that the use of proxy interviews may introduce some bias in our results. However, we have analysed the association between whether a proxy interview was Used for a particular respondent and the self rated health reported (using the dichotomous variable of good (ie very good or good on the 5-category scale) and bad (ie moderate, fair and poor) self-rated health. The chi-square test was not significant with a p value of 0.31. 3) It would be interesting for the reader to see the full distribution of selfrated health, in the total sample and in each gender group. Our response: We thank the reviewer for the suggestion. We have added a new table (please see Table 1) in our revised manuscript. 4) Combining all diagnoses into a yes/no variable results in a loss of a lot of information. The authors could consider using several categories according to the number of diagnoses reported and separating physical and mental health problems. Our response: We thank the reviewer for pointing this out. Yes, it could result in a loss of some information. As suggested, we have separated out selfreports of doctor-diagnosed illnesses into mental (comprising
anxiety/depression and insomnia) and physical illnesses (other self-reported illnesses). Please see the Table 2 in our revised manuscript for the new multivariate logistic regression analysis replaced with these new variables. 5) Discussion, first paragraph: the overall point is that there seem to be cultural variations in the subjective interpretation and use of the selfrated health response scale, as well as actual differences in the health of different populations. However, most studies have combined only the bottom 2 categories of SRH into a poorer health category and therefore it is impossible to compare their findings with those from the current study and to conclude that Singaporeans in general appear more positive in their ratings of health compared with other countries. Our response: When combining the bottom 2 categories of self-rated health into a poorer health category, the proportion of respondents was about 1.5%. We have amended this number in the first paragraph of the revised manuscript to reflect the proportion reporting their health in the lowest 2 categories, so as to allow comparisons between countries. 6) In general, and in relation to the previous comment, the authors may wish to cite other studies of SRH in Chinese population and refer to their findings: Yu et al 1998, American Journal of Epidemiology 147, 880-890. Ho 1991 American Journal of Epidemiology, 133, 907-921 Our response: We thank the reviewer for pointing out these studies in the Chinese population. We have included both references in our revised manuscript and also amended the paper to include a comparison between our findings and these studies. 7) The authors should discuss the limitations of their study (including, for example, the unstandardized translations to other languages). Our response: We have expanded the paragraph on the limitations of our study. These include the unstandardised translations into other languages. 8) Table 1 should also show columns for each gender group. Our response: We have expanded the Table 1 to show data by gender. 9) Page 12, second paragraph: income and education levels are still lower compared to men - yet in the adjusted analyses there was no relationship between these variables and SRH in women. Our response: We had used income and education as proxy indicators of status, but agree with the reviewer that the adjusted analyses do not show any relationship between income and education and SRH in women. We speculate however that other unmeasured elements of status may explain the poorer SRH in women compared to men. We have amended the relevant
paragraph in the revised manuscript and thank the reviewer for pointing this out. 10) Page 16 the stronger society pressures on women with regard to obesity are related to body image, not necessarily to health. Our response: We agree with the reviewer that societal pressures on women are related to body image rather than health. We have amended the relevant paragraph in the revised manuscript. Discretionary Revisions 1) Page 11, last paragraph: it may be clearer if the conclusion, which appear in the sentence before the last one in this paragraph (page 12) appears earlier in the paragraph and is followed by examples from different studies. Our response: We have moved the conclusion up the paragraph. 2) Sub-titles in the discussion section would be helpful (eg Gender differences on page 15, as a beginning to the section that discusses this issue.) Our response: We have added sub-titles to improve ease of reading of the Discussion section. Thank you once again for your kind consideration. We look forward to your favourable reply. Yours sincerely Dr Lim Wei-Yen