Data sources
PubMed database[15] was searched to obtain research volume of each countries from sub-Saharan Africa from January 1, 1981 to October 31, 2009. The numbers of HIV research articles indexed in PubMed was used as surrogate for total HIV research productivity. Articles originating from each country and published between 1981 and 2009 were generated by selecting the advanced-search option and then selecting the "publication date" field. Next, the "affiliation" field was searched for each country. The names of the countries were imputed in their different possible forms: Côte d'Ivoire and Ivory Coast for Côte d'Ivoire, and both Swasiland and Swaziland for Swaziland, for example. Some names of countries are also names of parts of other countries: Benin and Niger, for example are name of a place in Nigeria. To avoid errors arising from this, appropriate commands were used [i.e. (Niger [AD] NOT Nigeria)]. See additional file 1 for full search strategy. Though PubMed has been widely used for bibliometric analysis, it is important note that PubMed consist largely of English-language journals therefore possibly contributing to selection bias due to language barriers. In addition, PubMed do not represent all scientific and biomedical journals published. Many articles of biomedical importance appears in journals other than those included in the searched categories. Other limitations include the incorrect citation of origin for the authors, and definition of research production. By using the author addresses listed in the bylines of research articles, one can only identify countries and organizations where the authors were employed when the research was done or where the article was written, or both. Institutionally co-authored research articles co-publications are useful and tangible proxies of research involving African scientists and scholars.
Data on number of people living with HIV, adult literacy rate, gross domestic product (GDP), public expenditure on education (% of GDP), researcher and development researchers, number of higher institutions, number of indexed journal in MEDLINE, physicians (per 100,000 population), total expenditure on health, and private expenditure on health were obtained from the reports published by the United Nations Development Programs[16], World health Organization[17], and Joint United Nations Programme on HIV/AIDS[18].
Statistical analyses
Series of univariable and multivariable negative binomial regression models were used to explore factors associated with variation in HIV research productivity in sub-Saharan Africa. The study deployed negative binomial regression, rather than Poisson regression, because there was significant evidence of over dispersion. Negative binomial regression, employing a robust method and has been shown to be to address the failure of Poisson regression model in presence of overdispersion by adding a parameter that reflects unobserved heterogeneity among observations[19].
The univariate negative binomial probability distribution of Y is:
where Γ is the gamma function.
The factor change in the rate of HIV productive was calculated from:
where E(y | x, x
κ
) is the expected count for a given x, and E(y | x
κ
+ δ) is the expected count after increasing x
κ
by δ. The percentage changes in the expected HIV literature production count were calculated from:
To allow for weighted comparison among the countries of origin we calculated the ratio of the number of publications from a certain country to the GDP, total expenditure on health, expenditure on education, adult literacy rate, and number of people with HIV. Finally, the Pearson correlation analysis method was used to examine the association between HIV research productive and the selected indicators. For correlation analysis, indicators were log transformed to linearise these associations. Data were processed and analyzed with Stata 10 software (Stata Corp., College Station, TX, USA).