Bibliometric data in clinical cardiology revisited. The case of 37 Dutch professors

Table 1 shows all new numerical data as assessed by 31 December 2010 from the Web of Science of Thomson Reuters. This time, all documents were included apart from meeting abstracts, corrections and book reviews. Therefore, compared with the previous one, the present analysis also comprises letters (in general, a very minor subset of the total output). All parameters are self-explanatory, but for the sake of clarity, we will use the work of one of the eight more recently appointed professors in clinical cardiology, Dr. Suryapranata, to explain the calculation of the h-index. Figure 1 shows his 193 papers ranked from most cited to non-cited. The first paper was cited 979 times, the second 779 times and the third 439 times. A total of 24 papers were cited 100 times or more. Of his 193 papers, 26 remained uncited. In the inset of Fig. 1, both axes are restricted to 100 papers along the abscissa and 100 citations along the ordinate. Thus, the most cited 24 papers are missing in the inset, restricting them effectively to 76 papers. Paper number 46 was cited 47 times and paper 47 was cited 45 times. Therefore the h-index, indicated by the thick lines in the inset, is 46. Table 1 shows the names of the 37 professors in clinical cardiology in alphabetical order, followed by their affiliation in the left two columns. Then in column 3, the year of publication of the first paper appears. Compared with the data in our previous paper [1], we have amended some years of first publication because they were incorrect in our previous report [1]. Next, to the right, there are two framed boxes. In the left-hand box, all publications from each author between 1971 and 2010 are considered. The citation window was similar. Thus, a paper published by ‘Arnold AER’ in 1988 was followed from 1988 to 2010. In the right-hand box, the papers under consideration were restricted to the years 2005–2009. Their citation window was from 2005 to 2010. The rationale behind the right-hand box is to find a means to compensate the younger authors for the fact that they have had less time to create an oeuvre than the older authors. The rationale behind the citation window of 2005–2010 vs. a publication window of 2005–2009 is that papers published in 2009 had the chance to be cited in 2009 when published in January 2009, but not when published in December 2009. Within each framed box, we have assessed: the h-index based on all papers, the h-index of all first authored papers, the total number of papers, the total number of citations and the citations per paper.

At the right of Table 1, there is a column with ‘editing comment’. The first ‘y’ or ‘n’ indicates whether the author's work has been checked for less than his/her total number of initials. The ‘y’ or ‘n’ behind the slash indicates whether we were forced to edit the results by hand or not. This is necessary when there is another author with exactly the same family name and initials. Finally, at the bottom of the columns, the arrhythmic mean of the numbers are given. In case of the h-indices, there are two averages. ‘Average 1’ is the simple average of the 37 values. ‘Average 2’ is the square root of the average of the squared indices, which seems to us the appropriate way to determine it. At this stage, we present all data without further commenting on them, apart from the fact that we wish to underscore despite the fact that we are analysing the work of 37 individuals, these sets of papers may have a substantial overlap, in particular when the authors are working at the same university medical centre. The degree to which this overlap occurs also differs. From these data, it is not possible to calculate the values for each centre, let alone for all centres.

Finally, we underscore that there are now alternatives for the Web of Science of Thomson Reuters such as Scopus, Science Direct or Google Scholar. These databases have different systems of indexing. Therefore it is important that all groups and/or individuals under assessment are evaluated using the same database.

The h-index is a tool that can easily be explained and determined by individuals interested in their research performance. As could be expected, many alternatives have been formulated thereafter [15]. Table 2 shows the first 10 authors based on the h-index of all papers. The ranking is done by dividing the h-index by ‘scientific age’. At the top of the listing is Dr. Bax from LUMC with 3.28 h/year and an h-index of 59. Number 2 is Dr. Serruys with 3.15 h/year and an h-index of 107. Four authors from the LUMC and three from the Erasmus Center dominate this list. The averages at the bottom of the table (and in all further tables) are the average of all 37 professors, not just from the top 10. It remains undecided whether all 37 professors have an à priori equal chance to score the same h/year, because there is heterogeneity in citation frequencies even within clinical cardiology [14], which means that subfields as atherosclerosis, arrhythmias and congenital heart disease in fact require their own different reference standards.

When one compares the right with the left box, it is surprising that the h/year is 3.03 for all 37 cardiology professors during 2005–2009, vs. 1.58 during ‘all years’. Our observations therefore do not comply with Hirsch's expectation [2] that the h-index increases linearly over time. Concerning the individual aspects, we found that nine of the 10 professors who were leading over ‘all years’ were also leading over the period 2005–2009 (compare the left with right boxes).

Figure 2 shows the h-index and h/year of the work of Dr. Verheugt. The steady increase of the h-index over the years is obvious. The same is pertinent for the gradual increase of h/year, confirming that the normalisation to ‘scientific age’ cannot completely compensate for the advantage of ‘older’ scientists. Figure 3 shows a comparison of 28 professors who could be assessed both during 1971–2008 [1] and during 1971–2010 with an averaged period difference of 2.5 years. Thus, the ‘scientific oldest’ author, Dr. Simoons, was assessed over 37.5 years in the previous analysis [1] and over 40 years in the present years, whereas Dr. De Winter was assessed over a period of 14.5 years in the previous analysis [1] and over 17 years in the present one. Despite this small increase in time frame, the h/year increased from 1.43 ± 0.10 to 1.65 ± 0.12. The paired difference was 0.23 ± 0.04 (p < 0.0005). It follows that, as also shown in Fig. 2 for an individual, the normalisation of the h-index is an improvement, but does not eliminate the disadvantage of younger scientists.

An alternative to overcome the disadvantage of younger scientists is to limit the assessment to first-authored papers. This also makes it clear whether or not senior scientists remain active themselves other than by steering younger scientists or providing them with facilities. For senior (last) authorship, it is not possible to distinguish between authorship with a relevant intellectual contribution and ‘honorary authorship’. Besides, this can only be done completely manually. Such an approach will not be easily accepted by the scientific establishment, because it can be expected that it defends its position in the same way as in the rest of the society. Still, history has taught us that scientists are in general not at their top when they are over 45 or so. It will, for example, come as a surprise to many that younger reviewers of scientific journals produce better reports than older ones despite the difference in experience [16].

Another major advantage of focusing on first authorships is that in such a system each paper is only counted once. This may help to prevent the blurring that follows from the comparison of the scientific output of investigators with a large overlap in their papers.

Table 3 shows the performance of the 37 professors in terms of first authorships. At poll position is Dr. Serruys with 1.15 h/year with also, by far, the highest all time h-index for first authors (39). Five authors (Drs Von Birgelen, Van Belle, De Winter, Van Gelder and Zijlstra) appear on this list who were not on the h-index based on all papers. Drs. Serruys, Bax, Van Veldhuisen, De Feyter and Wilde are on both top 10 lists. For the whole group, the average was at 0.46 h/year, which returns ‘relevant productivity’ to a human scale.

When the focus was on the 2005–2009 period (right-hand framed box), only five authors, Drs. Serruys, Bax, Van Veldhuisen, Van Gelder and Wilde, remained in the top 10. The others in that top 10 were Drs. Voors, Heymans, Verheugt, Van den Berg, Van der Wall, Mulder and Waltenberger (ex aequo; not included in Table 3). The fact that both ‘older’ and ‘younger’ scientists could appear on such a list underscores its dynamic characteristics.

Although the h/year increased from 1.58 to 3.03 when ‘all years’ were compared with ‘2005–2009’ when it was based on all papers (Table 2), it decreased from 0.46 to 0.36 when it was compared between the same periods, but now based on first-authored papers only. Such numbers may provide the scientific arena with a return of the human scale in a period of galloping inflation. We have indicated previously that not much has changed since 1907–1916 when Lotka formulated his famous law [17], which says that the proportion of papers by authors publishing 1, 2, 3 and more papers follows 1/n ² in which n is the number of papers. Thus, of all authors, about 60% publish only one first-author paper in their career, 15% of authors publish two and less than 1% of authors publish eight or more first-authored papers. This situation was virtually unchanged in 1996 [18]. The enormous increase in publication of papers per author probably results almost completely from the fact that more scientists are recruited and from the strategic organisation of networks.

Table 4 shows the ranking of the professors in clinical cardiology based on the total number of papers. This list is dominated by the professors of the LUMC, all four being in the top 6. When the period was restricted to the publication years 2005–2009, eight of the 10 authors remained at the top of this listing focusing on ‘productivity’. The inflation of ‘productivity’ (see previous section) is underscored by the fact that all 10 professors more or less doubled their average number of papers over the years. This also follows for the averages at the bottom of Table 4 for all 37 professors. It was 11.0 over all years (including 2005–2009), but 21.3 for 2005–2009. Dr. Bax remained at the top with a miraculous 110.8 papers per year. In the top year 2009 he published 138 papers, which implies one paper each 1.9 working days. One may wonder where this will end.

Table 5 shows the top 10 based on the total number of citations. In this listing the Erasmus Center dominates the list at 1, 3 and 5. Also the position of the professors of the LUMC is prominent. Drs. Verheugt and Suryapranata from UMCN appear in this top 10. Eight out of 10 of these professors also appear in this top 10 when the period is restricted to papers published between 2005 and 2009. It is remarkable that for all 37 professors, the average number of citations obtained per year was the same when their whole publication period was assessed, or when it was restricted to 2005–2009. In both cases, there were just over 300 citations per year.

Table 6 shows the top 10 based on the number of citations obtained per paper. Here, there is no correction for the ‘scientific age’. The reason is simple. The number of citations for a paper increases during the first years after publication. It reaches a zenith and then wanes again. Thus, if one calculates the cumulative citations per year for a paper, the average increases also, reaches a zenith as well and then declines again. This creates a hybrid scenario even for the different papers of the same author.

Table 6 shows that this listing is headed by Dr. Arnold, one of the few professors not primarily affiliated with one of the eight university medical centres. His score of 77.45 citations per paper stems from the time that he was working at the Erasmus Center. This parameter remained high, primarily by publishing relatively few papers in recent years. This becomes clear in the right-hand box where the citations per paper have fallen to 13.83 and the ranking is 19 instead of 1. The same phenomenon is seen for the recently appointed Drs Suryapranata and De Boer with a decrease of citations per paper by two thirds. However, these numbers should be considered against the background that for all 37 professors, the average citation was 30.09 for the whole period and only 15.67 for the period 2005–2009 as cited between 2005 and 2010. The explanation is simple: the papers published between 2005 and 2009 had less time to accumulate citations than the papers published during all years. Therefore, for this parameter, the absolute numbers have little significance. The significance of the ranking is more important. In this listing the second rank of Dr. Simoons with 62.02 citations per paper is striking. His high ranking was consolidated when the ranking was based on the 2005–2009 period. It means that he has succeeded in focusing on work that was frequently cited. It is quite surprising that none of the professors affiliated with the LUMC appear in this top 10 either based on all years or on the 2005–2009 period. In the 2005–2009 period, the high positions of Dr. Peters (AMC) and the recently appointed Dr. Heymans (MUMC) are striking.

When the five parameters in this brief analysis were used to compile a top 10, it could have been possible, in theory, that each of these five top 10s would have consisted of the same 10 individuals. Alternatively, all 37 professors under assessment could have appeared one or more times. Instead, we found 22 out of 37 professors in at least one of the five top 10s. This underscores that there is no golden parameter. Two individuals, Dr. Serruys and the recently retired Dr. De Feyter, appeared in all five top 10s.

Quality assessment of scientific output remains a heavily disputed issue. We recommend, at this stage, to be very careful until better tools have been developed. The methods are not good enough and are too dynamic at the moment [14, 19] to arrive at criteria acceptable and fair for all involved.

We wish to state here that any future assessment system should (1) work with publicly available data and (2) should be understandable and (3) reproducible for scientists under assessment. The pivotal issue remains (4) a reference standard that is fair and objective. We can illustrate this with a very simple example directed to a specialty of one of us (AAMW) and one of our close colleagues, Dr. Mulder, in the Academic Medical Center. The h-index does not only apply to individuals, but also to journals and even to topics of research. Thus, when we select the six leading journals in cardiovascular science and define a publication period of 2000–2009 in the Web of Science of Thomson Reuters, we can automatically generate an h-index for topics. Selecting ‘Brugada Syndrome’ and ‘Marfan Syndrome’ would represent appropriate topics for part of the work of both scientists. The result is a h-index 59 for ‘Brugada Syndrome’ and a h-index 29 for ‘Marfan Syndrome’. When such large differences exist within what on the surface seems to be a relatively homogeneous field as ‘clinical cardiology’, little imagination is needed to see how such differences could work out for scientists who are active in fields as ‘robotics’ and ‘atherosclerosis’ in one and the same university medical centre. If we cannot find a solution for these problems, the combination of fundamental and clinical sciences into university medical centres may eventually end as a bad adventure when ‘science managers’ misinterpret or deny these issues.

Comparing citation frequencies of scientists in different (sub)fields may be regarded as assessing the skills of fishermen. In doing so, it is important to know (1) how many fishing rods are involved (money, co-workers, network), (2) how eager the fish are to bite (the number of references in the citing papers) and (3) how many fish are in the pond (number of scientists in the same (sub)field). In particular, this last point has not been addressed by the specialists in bibliometrics.

As in our previous analysis [1], the variability in h-index between individual professors in clinical cardiology was substantial (Tables 1, 2, 3). At this stage we repeat our hesitation to interpret the observed differences as differences in scientific quality.

We would certainly not publish an analysis of the scientific output of younger scientists at the beginning of their careers without more research. In this case, we thought that the information is of sufficient interest and the fact that the individuals have a completed career in the sense that they have reached the top of the academic ranks, played a role as well. The large interindividual differences as observed for these 37 professors in clinical cardiology should be an important warning for those who wish to compare scientists of different disciplines. To mention one pitfall, the citation frequency of clinical cardiology papers is about 40% higher than that of basic cardiology-oriented papers [14]. Within these categories the differences between subfields are probably even higher [14]. Ignoring such pivotal information may create a situation in academic hospitals where the quality of research of small fields is severely underestimated. It would be very counterproductive when high quality of research is misjudged by the use of poor quality indicators. This is a danger for smaller groups in smaller fields. Excellent work can attract citations from all other scientists active in the field and still score low because the number of scientists in the field is low.