Epidemiologic Perspectives & Innovations BioMed Central Analytic Perspective

Societies are facing challenges as the public health burden increases in tandem with population aging. Local information systems are needed that would allow a continuous monitoring of the incidence and effectiveness of treatments. This study investigates the possibilities of routinely collected administrative data as a data source for hip fracture incidence monitoring in Finland. The study demonstrates that a straightforward use of register data results in biased estimates for the numbers of hip fractures. An interpretation of hip fractures from the population aging point of view offers an alternative perspective for hip fracture incidence calculation. This enables development of a generalizable method for probabilistic detection of starting points of hip fracture care episodes. Several risk factor and risk population extraction techniques required in register-based data analyses are also demonstrated. Finally, it is shown that empirical evidence suggests that hip fracture incidence is proportional to population level disability prevalence. In conclusion, Finnish administrative data makes it possible to derive data for rather detailed population level risk factor stratification. Certain limitations of register-based data can be partly avoided by synthesizing data-sensitive methodological solutions during the analysis process.


Introduction
The mission of this Journal includes publishing "epidemiology-based policy analysis, critical analyses of the field and its practices, or various contributions to methodology, philosophy, or other perspectives on the field" [1]. As is true for most journals, Epidemiologic Perspectives and Innovations wishes to publish work that is influential at many levels, including clinical practice, health policy, and future research. One objective measure of the influence of a publication on future research is the frequency with which the study is cited in subsequent publications. "Citationology", a term coined by Garfield, refers to "the theory and practice of citation, including its derivative disciplines citation analysis and bibiometrics" [2]. Citationology has been examined in many fields, ranging from sociology [3] to health sciences [4]. In these studies, much attention has been focused on the large body of literature that is uncited and somewhat less on the characteristics of the relatively few journals in which a large proportion of all cited work is published [5].
In a two-article series using data from the Institute for Scientific Information (ISI), Hamilton found that approximately 46% of papers in medicine remain uncited 5 years after publication [6,7]. However, as Pendlebury noted in a letter to the editor, this may be an overestimate because Hamilton's statistics include every type of publication, including obituaries, editorials, abstracts, letters, and other marginalia which, together, represent about 27% of all items indexed [8].
In another study, Callaham et al. investigated journal prestige and other characteristics associated with the citation of studies in peer-reviewed journals and found that the impact factor was the most important predictor [9]. Factors related to citation frequency differed depending on the field of enquiry, however.
In this study, we sought to determine if author, institution, country, journal, or topic are associated with the number of times a publication is cited. To test the robustness of our findings we chose to examine these relationships in 2 highly contrasting areas, child injury prevention and coronary artery disease (CAD) prevention, both of which were viewed from an epidemiological perspective.

Materials and methods
We used outcome-based sampling to examine this relationship and searched the ISI Web of Science ® , including the Science Citation Index Expanded™ (SCI EXPANDED™), the Social Sciences Citation Index (SSCI), and the Arts & Humanities Citation Index (A&HCI). For child injury prevention, we used the search terms "child*" and ("injury" or "accident") and "prevention", and for CAD prevention, we used ("coronary artery disease" or "coronary heart disease") and "prevention". We performed this search in March 2007. All searches were limited to full articles containing original research involving human subjects and to articles published in English between 1998 and 2004. We examined the results year by year. To minimize the potential effects of different latency periods between publication of the original manuscript and publication of the citing manuscript, we limited the search to articles published between 1998 and 2004, thus providing at least 2 years for the most recent citing manuscript to be published.
We assumed that the number of citations is affected by the time since publication, e.g., that an article published in 1998 is likely to have been cited more frequently than one published in 2004. To account for this, we calculated the citation rate using the following formula: Consequently, if an article published in 2000 was cited 300 times, its citation rate would be 300/(2005-2000) = 60 citations per year. To minimize bias, we also used frequency-matching, selecting the 36 most frequently cited and 36 most infrequently cited articles for each year for each topic area (child injury prevention and CAD prevention). With 36 highly-cited and 36 infrequently cited articles per topic per year, our final sample had 252 highlycited and 252 infrequently-cited articles examining child injury prevention and 252 highly-cited and 252 infrequently-cited articles examining CAD prevention.
For each article we recorded the following characteristics: authors, journal name and type, topic, institution, country, and year of publication. In addition, we obtained the circulation for each journal from Ulrich's International Periodicals Directory. Because previous circulation data were not available, we used current circulation as a proxy, and categorized the results as low (<18,000), medium (18,000 -50,000), or high (>50,000) circulation. Impact factors for 2001 were obtained from ISI Web of Knowledge. A journal's impact factor represents the mean number of times an article published in the last two years was cited [10]. We categorized impact factors as low (<2.7), medium (2.7 -9.0), or high (≥10.0). These 2001 impact factors were used as a proxy for the average of each journal's impact factors over the study period.
Among publications examining child injury prevention, we classified articles according to the following topics: burns/fire, education, guns, motor vehicle, playground/ sporting, or other injuries. Among those examining CAD prevention, we grouped topics into medications, new technologies, risk factors, or other issues. Statistical analyses consisted first of a descriptive phase where we provided the distribution of the following characteristics: authors, institution, country, topic, and journals by name, type, circulation, and impact factor. Nominal data are presented as counts or proportions, and continuous data as means ± standard deviation (SD). In a subsequent analytic phase, Fisher null hypothesis tests were used to examine the association between whether a publication is highly or infrequently cited and the following factors: country, topic, and journal name, type, circulation, and impact factor. Nominal data were compared using Chi-square or Fisher's exact tests, as appropriate, and continuous data using student t-tests. We also used the stratified (Cochran-) Mantel-Haenszel to test the association across impact factor strata. All statistical analyses were performed using SAS 9.1 (SAS, SAS Institute, Gary, NC).

Author, institution, and country
The CAD prevention articles involved over 2,500 authors whereas the child injury prevention articles involved 1,700 authors. Over 90% of authors appeared on only 1 publication. Most institutions of origin also only appeared on only 1 publication during the study period. Institutions with more than 2 publications are listed in Tables 1 and 2. The Centers for Disease Control had the greatest number of highly-cited publications on injury prevention, whereas Harvard University produced the most highly-cited research papers on CAD prevention.
Studies conducted in the United States were more likely to be highly-cited than those conducted in other countries. This relationship was statistically significant for both topic areas (Tables 3 and 4).

Journal
Highly-cited papers were most likely to be published in journals with large circulation or high impact factors, and circulation and impact factor were highly correlated (r = 0.77). Thus, among CAD prevention publications, highlycited articles were more likely to be published in Circulation, the Journal of the American College of Cardiology, the Journal of the American Medical Association, the Lancet, or the New England Journal of Medicine rather than other journals (Table 3).
For injury prevention papers, highly-cited publications appeared in the Archives of Pediatric and Adolescent Medicine, Burns, Injury Prevention, the Journal of Trauma, and Pediatrics (Table 4). Thus, in both categories, highlycited publications appeared in journals with medium or high circulation and medium or high impact factors (Tables 3, 4) and infrequently-cited papers were in journals with low impact factors or low circulation.

Topic
Topic was also associated with citation frequency only among CAD prevention papers. Highly-cited articles were more likely to discuss risk factors, whereas infrequently cited papers addressed other, general topics (Table 3). This association persisted after adjusting for impact factor (p = 0.04).
For child injury prevention, there was no significant association between topic and frequency of citation (Table 4). Infrequently-cited papers may be more likely to examine burn prevention than highly-cited ones, but this relationship did not reach statistical significance and remains after adjusting for impact factor (p = 0.68).

Discussion
Perhaps not surprisingly, the journal, reflecting in part impact factor and circulation, was most strongly associated with the frequency with which a publication was cited. This was particularly evident among publications examining CAD prevention, where almost 80% of highly- cited articles were in medium-or high-impact journals compared with 12% of infrequently-cited publications.
For CAD, we also found topic to be associated with frequency of citation and to the country where the research was done. For both CAD and injury prevention they were far more likely to come from the United States.
The factors found to be related to frequency of citation are themselves highly correlated, and these correlations must be considered when interpreting these results. For example, it is likely that journal of publication lies in the causal pathway because high-impact journals more often publish articles about interesting or timely topics. Consequently, adjusting for intermediate variables such as journal, impact factor, or circulation is likely to result in an underestimation of the effect of topic on citation frequency. In addition, impact factor is derived from how often a journal is cited. It is thus difficult to interpret the importance of the association between impact factor and future citation. Furthermore, adjusting for impact factor, which is highly correlated with frequency of citation by definition, may attenuate the observed effect of topic. However, similar results were obtained when adjusting for journal circulation instead of impact factor. This indicates the importance of choosing the 'best' journal (high circulation, high impact, or both) if authors hope to influence readers by virtue of the frequency with which their work is likely to be cited by others.
Only a small number of studies in the epidemiological literature have examined publication characteristics associated with citation frequency. The most prominent of these was performed by Callaham et al. [9]. They found impact factor to be the most important predictor, as did we. Additional important predictors were subjective newsworthi-  In another study, Callaham et al. examined the citation rate among articles submitted to emergency medicine research journals compared with non-emergency medicine journals [11]. They found that the average citation rate among the emergency medicine journals was 2.04 times per year whereas the citation rate among similar articles published in non-emergency medicine journals was at least twice as high. Fifteen percent of articles published in emergency medicine journals were never cited, compared with only 5% of those published in non-emergency medicine journals. This finding could be attributable to the fact that the mean impact factor of the nonemergency medicine journals was substantially higher than that of the emergency medicine journals (4.3 vs 1.5), which further highlights the importance of impact factor as a predictor of the frequency of citation.
Previous citation studies have also examined the rates of papers that are never cited [12]. Schwartz found that 46% of all publications in medicine (including marginalia) were never cited by other authors [12]. Considering only full articles, the proportion was 22% -similar to rates from other physical sciences, including Physics (47% among all publications, 17% among articles only), Biological Sciences (41% and 19%, respectively), and Mathematics (55% and 26%, respectively). The proportion of uncitedness in these fields is substantially lower than those of other disciplines, including the Humanities (98% among all publications, 93% among articles only) and Social Sciences (75% and 48%, respectively).
The work conducted by Schwartz also highlights a key challenge in citationology -the definition of the 'population at risk'. It remains unclear whether all publications in journals catalogued by the ISI should be considered 'at risk' for citation or whether analyses should be restricted to original articles. Furthermore, given the imperfect sensitivity of literature search tools, even well-defined searches will likely produce incomplete results. As a result, it is difficult to identify the entire 'population at risk' of being cited or to calculate measures such as the risk of citation.
These difficulties in identifying the 'population at risk' also prevent the use of Kaplan-Meier survival analysis which could provide valuable information regarding time to citation -a dimension of likely relevance for many authors.
Another important challenge in examining citations is the effect of time. Calendar time can affect the citation of an article in three important ways. First, there is the potential effect of partial years. It is unlikely that an article will be published on January 1 st and equally unlikely that a literature search will be conducted on December 31 st . Consequently, both the year of publication and the year of search are, in fact, partial years, the effects of which need to be considered when interpreting the results. To minimize the effects of these partial years, we frequencymatched on year of publication. Second, there is a latency period between the decision to cite an article and the publication of the citing article. This latency period can be highly variable, depending on the number of times the article is submitted, different review times, and the duration of the 'in press' period. Although the increasing popularity of open access and electronic journals likely will shorten this period, a minimum, unavoidable latency period will always remain. In the present study, therefore, we restricted our analysis to papers published between 1998 and 2004, thus allowing at least two years for the citing manuscript to be published. Third, the citation trajectory of most articles is concave-down ( Figure 1); i.e., most articles experience a period of increasing citation followed by one of decreasing citation. As such, it is perhaps too simple to just examine rates of citations per year. Calendar time-citation interaction is particularly important when focusing on specific citation rates.

Limitations
Our study has several limitations. First, to detect important differences for some variables (e.g., author, institution), even this relatively large sample is insufficient. It is also possible that the significant results may be the result of type 1 errors.
Second, although our results are generally consistent for CAD prevention and child injury prevention, the generalizability of the findings is uncertain. Additional studies involving other epidemiologic topics will be needed to determine how universal these results are.
Third, we used 2001 impact factors and assumed them to be roughly the average of any journal's impact factors over the study period. Although it would have been preferable in a design sense to have used 1998 impact factors, because the impact factors of some journals have risen sharply over the last 12 years [13], these may not be representative of later values. However, with a maximum of only 4 years between publication and the 2001 midpoint and the categorization of impact factor as low, medium, or high, it is unlikely that the use of 2001 impact factors as a proxy introduces an important bias.
Fourth, we used current circulation data as a proxy because earlier circulation data were unavailable. This is not ideal, but journal subscriptions do not greatly fluctuate over a 7 year period and it is especially unlikely that a journal would move from one circulation category (low, medium, or high) to another over this time. Also, due to the presence of institutional subscriptions, available subscription data likely underestimate the number of readers of a given journal.
Finally, as discussed previously, there are the potential effects of calendar time that we have tried to minimize by frequency-matching and by incorporating a minimum latency period into our study design.

Conclusion
Journal, impact factor, and circulation were the factors significantly associated to the frequency of citation of epidemiologic publications. These associations were present for publications addressing child injury prevention and those examining CAD prevention. Topic also appeared to be associated with citation frequency among CAD prevention articles.
Publish with Bio Med Central and every scientist can read your work free of charge