Abstract
Backgrounds Traveler’s thrombosis is well-known even among non-professionals as “economy class syndrome.” However, controversial discussions have remained. Clark et al. mentioned no solid evidence in their article “Long-haul travel and venous thrombosis: What is the evidence?” (1). Also, the status of recent COVID-19-related thrombosis research seemed like that of the traveler’s thrombosis. We hypothesized there might be something overlooked behind those strange situations.
Methods Since ordinary review methods (e.g., systematic review or meta-analysis) had already been conducted (2–7), we focused on reviewing a “previously published” “chart.” Also, we developed a novel “review method” for the meta-regression analysis result. We applied those to some previously published and well-known data.
Results We newly found an approximately 28 days cycle of thrombosis onset over several weeks after travel in a figure (Cannegieter et al., PLoS Med. 2006, in the top 25% of all research scored by Altmetric) (8). Also, we found an eighteen-day cycle of thrombosis onset in another chart (Kelman et al., BMJ. 2003, in the top 5% of research) (9). In COVID-19, we newly extracted subgroup patterns in a scatterplot (Troponin T and NT-proBNP) that applied simple linear regression analysis (Guo et al. JAMA Cardiol. 2020, top 5% research) (10). Also, these subgroups have already appeared in the cardiomyopathy study (Budnik et al., Int J Cardiol. 2016) (11).
Conclusions There has been a widespread belief that in-flight environments cause thrombosis. Also, the phenomenon of the onset of thrombosis after over a few weeks to two months since the travel has been explained, such as the development of blood clots occurring on a cabin, but the dislodge of a thrombus occurs after the travel (12). However, from cyclic patterns of thrombosis onset, it is more reasonable to explain that the “travel” and “high-risk period of thrombosis with Oral Contraceptives (OC) use (within three months from starting OC)” coincided with the travel (e.g., honeymoon and starting birth control OC). The influence of the in-flight environment may be less than we had previously thought. Although simple linear regression analysis has been used in studies of cardiovascular biomarkers, including the case of COVID-19, there seems to be a complex scatterplot structure that is unsuitable for simple linear fitting. Considering our findings, in a literature review, a pattern on a chart should be given more paying attention.
What is already known on this topic Strange controversies have remained in thrombosis-related fields (traveler’s and COVID-19-related thrombosis). Traveler’s thrombosis sometimes occurs over two months after leaving the risky in-flight environment. This phenomenon has been explained that the thrombus is formed in a cabin but dislodged after (12). Oral contraceptives (OC) were recognized as aggravating factors (13). COVID-19, although this is the first time humans have faced SARS-Cov-2, there has been a long research history on infectious-associated thrombosis.
What this study adds We hypothesized something might be overlooked and focused on reviewing a “previously published” “chart.” We found thrombosis onset cycles. These findings may imply thrombosis after travel is caused by overlapping the travel period and the initial period of taking OC, which is the high-risk period for thrombosis. In COVID-19 cardiovascular biomarker studies, there seems to be a complex scatterplot structure that is unsuitable for usually used simple linear fitting. In a literature review, a pattern on a chart should be given more paying attention.
Main text
Introduction
Research question and hypothesis
Thrombosis research has a relatively long history in medicine. In the 19th century, Virchow’s triad, named after Rudolf Virchow (1821-1902), was a principle for thrombus development (“Hypercoagulability,” “Stasis,” and “Endothelial injury”)(14, 15), had already been established.
However, controversy has remained in “traveler’s thrombosis (travel-related thrombosis)” (Supporting discussion 1: history of traveler’s thrombosis)(1-8, 13, 16-24). Clark et al. noted no solid evidence in their article “Long-haul travel and venous thrombosis: What is the evidence?” (1). Despite traveler’s thrombosis being well-known among medical professionals and the public as “economy class syndrome,” remaining controversial seemed strange. Also, considering some controversy (14, 15, 25–29) about recent COVID-19-related thrombosis research, the situation seemed like that of the traveler’s thrombosis (Supporting discussion 2: research situations of COVID-19-related thrombosis).
We hypothesized something might be seriously overlooked behind those strange situations.
Making a “review strategy” to find out something overlooked
However, trying ordinary review methods (e.g., systematic review or meta-analysis) again seemed impractical to extract something overlooked because those types of the study had already been conducted (2–7). So, a novel strategy seemed to be needed.
Firstly, we focused on reviewing “figures.” Usually, numerical data in tables within each piece of literature are evaluated in a meta-analysis. However, figures created by each researcher are generally not. Also, seeing the chart is more difficult than we thought (30).
Recently, there has been software that detects research fraud in the figures submitted for journals (31), but it does not discover “overlooked findings.” Also, although raw data images, such as CT or MRI images, were analyzed, there is no awareness of new information extraction from the “previously published” “chart.” So, it seemed helpful for novel method development that reviewing previously published charts to find unknown information.
Secondary, we focused on geometry. Although statistics usually have applied to medicine (e.g., p-value), approaching from a different direction seemed helpful when ordinary methods do not work. We finally made our original strategy, which was extracting overlooked knowledge from the “previously published” “chart” by “geometrical viewpoint” (Fig. S1, a-c).
Developing a novel “review method” for meta-regression analysis result
Considering the above strategy, we focused on a figure that showed a result of meta-regression analysis (Fig. S1, b)(Chandra et al., Ann Intern Med. 2009;151(3):180-90., Figure 3) (3). In this figure, we saw a geometrical shape (layered hyperbolic shapes) that seemed to be formed by two subgroups (Fig. 1, Fig. 3, a). Also, we conceived a methodology, which was a manually executable kind of pattern recognition, searching layered hyperbolic patterns in a figure that shows a result of meta-regression analysis to detect latent sub-groups. Our basic idea is that a confidence interval length for an odds ratio is inversely proportional to the “potential patient population size” (see Fig. S2, d-f) (Details in Methods).
a, A two-way cross-tabulation (contingency table) for calculating an odds ratio (OR). b, Calculation of OR and its 95% confidence interval. c, Histogram of the data following a binomial distribution and a plot of the inverse values of the square roots. d, The mathematical formulas for the hyperbolic shape. Note that the OR is not logarithmic, and only the length of the error bar is logarithmic (see the upper left position of panel d). We generate binomial distribution data because normal distribution could be approximated by binominal distribution. Around the center of distribution (within ±2SD of the mean), the parabola was a good fit for the points calculated from the simulated data. Also, from around the over ±2SD to ±3SD range, the parabola is positioned below the points. Although a catenary curve (dashed curve) is more likely suitable for fit, the number of patients who fall into the categories at both ends is slightly larger than the expected number. Usually, data placed over the range of ±3 SD from the distribution center are summarized as “greater than” or “less than.” For example, in the case of 10 hours is the center of the distribution, the time category might be following sets such as “less than” 5 hours, 6-8 hours, 9-11 hours, 12-14 hours, and “more than” 15 hours. The value of 1 over root N (1/sqrt(N)) becomes smaller. So, using a parabola seemed to be based on the reality of the data handling method in the medical and biological fields.
Our report in this article
This article shows the results of applying our strategy and method (the overall study framework is demonstrated in Fig. 2).
We applied our concepts to travelers’ thrombosis and COVID-19. These were similar in research history (see Supporting discussion 1: history of traveler’s thrombosis & Supporting discussion 2: research situations of COVID-19-related thrombosis). In the case of traveler’s thrombosis, we analyzed a dataset collected by Chandra et al. and Philbrick et al. (3, 4). In the case of COVID-19, we analyzed a dataset collected by Matsushita et al. (3). Flow charts show data accept or reject flows. To avoid duplicating data, Matsushita et al. selected only eight studies data (25 studies included in initial web Figure 5) (41).
Methods
Deriving the exact formula directly was difficult. So, firstly we focused only on the U-shaped curve. The length of the confidence interval for the odds ratio is inverse proportional-like to the size of the potential patient population, which onset thrombosis by exposure to risk factors (see Fig. 1). Although, there are multiple possibilities (e.g., 1/N, 1/N2, or 1/sqrt(N)), regarding the form of equations (1) and (2) below, we intuitively predicted that it would be approximately equal to the inverse of the root (equation (3)).
To be precise, the ability of patient recruitment, in other words, the “sample size in each study,” affects the confidence interval length.
It is similar to Hooke’s law on spring (32), which states that the pull-back force of spring is proportional to the length of the pull. In this case, the power also depends on the spring type (e.g., thick spring or thin spring). Thus, the formula is as follows, with spring constant (k) depending on the kind of spring (equation (4)). Similarly, equation (3) express equation (5) precisely.
However, we speculated that the effect of the differences in sample size could not be a practical problem for two reasons.
The first reason is that stratified analysis, often performed in clinical studies, can be a force in reducing the differences. The second reason is that two types of boundary conditions in the hyperbolic shape fitting, which are the “S-curve” and the “error bar length,” reduce the degree of freedom in searching and increase the possibility of finding the desired shape.
Considering the numerical experiment (see Results - Numerical Experiment), we decided on the equations for the S-shaped curve, the upper end of the confidence limit, and the lower end of the confidence limit (equation (1), (2), and (3)). Note that in the below equations, the coefficient L is generally set as 1. So, we used this equation under the condition of L=1 unless there is some reason.
The whole process, including, data review, regression analysis, and tools for the analysis, is described in the supplementary information (Supporting information: Method details).
Results
Numerical Experiment
Usually, a sigmoid function is used for curve fitting to a dose-response relationship and outline of the cumulative distribution. Also, the curve expressing the end of a confidence interval is a sum of the S-shaped and U-shaped curves (see Fig. 1).
To decide on a formula to express a U-shaped curve, we performed a numerical experiment. The result indicated that a parabola could approximate a U-shaped curve (Fig. 1, c & d). We constructed three mathematical formulas for fitting hyperbolic patterns to data (see Fig. 1 and Methods).
Analysis 1A (dataset: Chandra et al.) (3)
Before redrawing the figures of the meta-regression analysis, we reviewed the cited four studies (Martinelli et al., 2003; Parkin et al., 2006; Cannegieter et al., 2006; Kuipers et al., 2007) (8, 13, 17, 33). The main objective of Chandra et al.’s research (3) was estimating a risk, but the main of our analysis is detecting the subgroups. So, we are considering the possibility of some problems with inclusion in our study, even if there were no problems in Chandra et al.’s research.
In this assessment, two study data (Martinelli et al., 2003; Cannegieter et al., 2006) (8, 13) were judged inappropriate and excluded from our analysis. Cannegieter et al. and Martinelli et al. used a unique research design, which uses the patient’s partner as a control (Martinelli et al.: friend and partner). In the introduction, we described that although the properties of each study (the ability of patient recruitment, sample size) disrupt the effect of the size of the potential patient population, the stratified analysis could reduce the disruption. However, the impact of adding a unique study design was unclear.
Additionally, the odds ratio reported by Parkin et al. (33) was suspected to be affected by the misplacement of the cells in the cross table for calculating the odds ratio. Still, qualitatively, it could be used (see Fig. 2, Table S1, Table S2, and Supporting information: Method details).
In the residual dataset (Parkin et al., 2006 & Kuipers et al., 2007)(17, 33), layered hyperbolic patterns appeared, and applied regression analysis with the above three formulas. The inflection points of the S-curves (center of distributions) were 7.1 hours and 11.8 hours, respectively (Fig. 3, b).
a, Data review and meta-regression analysis using the same method as Chandra et al. Ann Intern Med. 2009; 151(3):180-190. Figure 3 (3). We added information from original studies and our hypothesis to the previously published form, such as latent distribution. This figure is regarded as a re-drawn figure from Chandra D, Parisini E, Mozaffarian D. Meta-analysis: travel and risk for venous thromboembolism. Ann Intern Med. 2009 Aug 4;151(3):180-90. doi: 10.7326/0003-4819-151-3-200908040-00129. Epub 2009 Jul 6. © 2009 American College of Physicians. Adapted with permission. The original figure has been shown on the American College of Physicians website, which links to PubMedR○ (https://pubmed.ncbi.nlm.nih.gov/19581633/). b, grouping the data and hyperbolic shape fitting. The dots represent the odds ratio with error bars of the confidence interval are the adjusted odds ratio described in the four studies cited by Chandra et al. Since Cannegieter et al. did not describe values of confidence intervals, we calculated the values from the case-count data. We also calculated the odds ratio at the time point of short travel in the research reported by Martinelli et al. Chandra et al. did not display it might be due to the no adjusted odds ratio initially described. Our result of a meta-regression analysis was slightly different from the straight line drawn by Chandra et al. might be due to the software used by Chandra et al. (STATA) being different from the one we used (R and R package “metafor”) or they might conduct a meta-regression analysis reversing the front head and the front side of the cross-tabulation. It was impossible to fit S-shaped curves for the points excluded from the search for hyperbolic shapes (Martinelli et al. & Cannegieter et al.) due to the small number of data or the form of the point sequence. However, if we manually fit S-curves, the two inflection points would appear close to the positions of the hyperbolic shapes’ inflection points.
Additionally, it seemed that S-curve could be fitted to the group of points reported by Cannegieter et al. and Martinelli et al., respectively. Also, those of inflection points would position at positions of less than 10 hours and more than 10 hours. This result seems consistent with the two hyperbolic shapes fitting (Fig. 3, b, orange, and green dots).
Notably, we observed an overlooked cyclic pattern of thrombosis onset in the figure reported by Cannegieter et al. (2) during the above data review process. The cycle period was 27.4 days (Fig. 4, a-e). The wave of about 28 days consisted of a gradually decaying wave and a monotonically decreasing function (Fig. 4, e). The wave disappeared after about three months (90 days). The monotonically decreasing part estimated that it continued for over 90 days after the travel (Fig. 4, f). Cannegieter et al. (2) did not mention this wave pattern.
a, A bar graph showing the relationship between weeks after travel and the number of thrombosis onset in a figure reported by Cannegieter et al. (8). b, Re-expressing as a two-dimensional bar graph avoiding the three-dimensional representation. c, Applying a damped wave by non-linear regression analysis. d, Extraction of damped wave part by subtracting the monotonic decrease function. e, Dividing into 2 sub-parts areas by the envelopes of the damped wave that touch the lower parts of the wave. f, Enlarged and added explanation of the ratio of the area. Cannegieter et al. (8) used the data up to week eight. The ratio of sub-parts 2 to the total number of patients (ratio of S2 to the area of S1 + S2) was 30.0%. In the integration for the infinite interval, it was 30.8%. Peak shifts of the wave (peak position of the wave shifted from the original to the other) appeared when comparing panels d and f due to putting by regression curve located in the center (c + d). Panel a was re-used from Cannegieter et al. PLoS Med. 2006; 3(8):e307. Figure 1. https://www.ncbi.nlm.nih.gov/labs/pmc/articles/PMC1551914/figure/pmed-0030307-g001/ Copyright © 2006 Cannegieter et al. Creative Commons Attribution License. In 2006, the Creative Commons Attribution 2.0 Generic, License (CC BY 2.0) was available. https://creativecommons.org/licenses/by/2.0/
Analysis 1B (dataset: Philbrick et al.) (4)
Considering that only a few pieces of data were available for the first analysis, we performed a similar analysis using another dataset gathered by Philbrick et al. for a systematic review. As a preparation, we conducted a data review to confirm eligibility for our regression analysis (see Fig. 2). The dataset did not contain risk ratio data. In this case, hyperbolic shapes could not be fitted (for the details, see Supporting information: Method details). So, only S-curves were applied. This analysis found two S-curves in the stratified datasets by pulmonary embolism (PE) and deep vein thrombosis (DVT). The inflection point of PE was about 9.2 hours, and the point of DVT was about 12.1 hours, respectively (Fig. S3, b), which were similar to the first analysis.
Additionally, we found that unrecognized periodic pattern in the data reported by Clérel and Caillard (34) during the data review, which was suspected to be the effect of circadian rhythm (Fig. S4, c).
We also found a cyclic pattern after travel in the data reported by Kelman et al. (9). However, the period of the wave was 17.9 days, and the wave was less clear than the wave observed in the data reported by Cannegieter et al. In the waveform, there was a raw-risk period just before 90 days. Also, in this data, there was an uptrend (Fig. 5).
a, A thrombosis onset distribution reported by Kelman et al. (9). b, The regression curve is located in the center of all average points (the points express the average of thrombosis onset during seven days). c, Curve fitting to the residual data of the regression curve in panel b. d, Application of damped wave function and adding interpretation of the results assuming some women started taking oral contraceptives (OC) in the timing of travel. The observed cyclic pattern was relatively unclear than Cannegieter et al.(8) (see Fig. 4). Noteworthy, the thrombosis onset downed in the days before 90 days, and it seems to be the scheduled withdrawal period of OC use. This figure was re-used and re-drawn from Kelman et al. BMJ. 2003; 327(7423):1072. Figure 1 https://www.bmj.com/content/327/7423/1072#F1 Copyright © 2003 BMJ Publishing Group Ltd. All rights reserved. The BMJ permission team thankfully confirms this figure adaptation. Also, we obtained permission to re-use.
Furthermore, we found a correlation between the stepped stages of increasing PE patients and the issuance of the guidelines (35–38) for preventive hormone replacement therapy (HRT) to a menopause woman in a figure reported by Clérel and Caillard (34) (Fig. S5). Especially the thrombosis onset in Paris airports has increased since the late 1990s (Fig. S5).
Analysis 1B-related additional analysis
Trend analysis and history of sex ratio change
We performed the trend analysis of the literature to assess the generalization on the increasing number of patients at Paris airports reported by Clérel & Caillard (34) (Fig. S5). We searched the literature in PubMed (keywords: “Economy Class Syndrome,” “Traveler’s Thrombosis,” and “Travel-related thrombosis”). The number of references increases rapidly after the 2000s (Fig. S6).
The difference between the above result of trend analysis and the result in the data reported by Clérel and Caillard (34) can be explained by the difference between publication day and onset day. So, there is no practical difference between the two results.
The literature search found that two articles (Symington & Stack, 1977, Cruickshank, Gorlin, & Jennett, 1988) (12, 39) contained each patient’s details. We already obtained each patient’s data from the other two articles in analyses 1A and 1B (Clérel & Caillard, 1999, Parkin et al., 2006) (33, 34). So, we confirm changing the sex ratio using those data to assess the effect of HRT. We found that the rate of women in the PE data gathered from the above four studies (12, 33, 34, 39) gradually increased but was most pronounced during the late 1990s (Table 1).
Literature review targeting early literature and analysis focusing on men
We also conducted a literature review targeting early literature because we strongly questioned the “economy class syndrome” concept from our findings in Analysis 1A and 1B. We found that the word appeared in Symington & Stack, 1977 (39). However, judging from the citation history checked on the Europe PMC website (https://europepmc.org/), the article by Cruickshank, Gorlin, & Jennett, 1988 (12) might influence the word as evangelists. The paper was a “personal paper,” a type of article category of the Lancet, and the authors prepared case reports of their own experiences.
Dr. Cruickshank was middle-aged (12). Usually, the mid-life crisis is concerned around this range of age. Dr. Jennett was a dean of medicine at Glasgow (40). Generally, heavy psychological stress based on heavy social responsibility is concerned. So, we hypothesized that men’s life event-related psychological stress could be one of the risk factors (see Fig. S7, b).
To assess our hypothesis, we merged data from the above four studies (12, 33, 34, 39) and made a bar chart for 1-year. The onset of PE was concentrated in certain age stages, such as around 30, the 40s, late 50s, and early 70s (Fig. S7, e).
We performed the same analysis in the women’s data except for data reported by Clérel & Caillard (34). Although the peak range was 65 to 70 in the data of Clérel & Caillard (34) (Fig. S4, d), PE was concentrated in the 40s to 50s (menopausal stage) in this analysis (Fig. S8).
Additionally, we found an inverted-U-shaped (mountainous) relationship in the scatter plot: age on the horizontal axis and air travel time on the vertical axis (Fig. S10, d). However, it was possible to fit a straight line as an up-right linear relationship or a down-right linear relationship(Fig. S9, a, and Fig. S10, a, c).
These analyses were also performed on the female data reported by Clérel & Caillard (34), and we found a U-shaped relationship (Fig. S9, c).
Analysis 2 (dataset: COVID-19) (41)
We decided to apply our ideas to the current complex COVID-19 pandemic problem to find some overlooked things because there was some controversy (14, 15, 25–29) about COVID-19-related thrombosis. This situation was similar to that of the traveler’s thrombosis, and we started this analysis with the inspiration from the famous mathematician Polya who stated that solving a similar problem helped us solve a more complex problem (42).
As a note, we described that stratified is necessary for the introduction section, but clinical studies of COVID-19 did not perform it. However, Stratification seemed unnecessary because the size of the studies was not huge and roughly the same due to no mega trial during the pandemic.
One of us (KK) searched for the layered hyperbolic pattern using the search engine Google and extracted a figure reported by Matsushita et al. (41) (Fig. 6, a-c). After evaluating eligibility (see Fig. 2), patterns that allowed to be fitted hyperbolic shapes appeared (Fig. 6, d-f, & g-i).
a, A figure reported by Matsushita et al. shows the relationship between the severe and non-severe groups (41), which shows age difference on the horizontal axis and odds ratio (OR) or hazard ratio of hypertension on the vertical axis. To evaluate potential confounding for relative risk by age, Matsushita et al. conducted meta-regression analyses based on the assumption that there was the possibility of confounding by age in the case that the study with a larger age difference has a higher relative risk (41). b, Diabetes. c, Cardiovascular disease (CVD). d-f, Comparisons of error bars, which show 95% confidence interval (C.I.) s. It corresponds to the upper figure. g-i, Hyperbolic patterns were fitted to the OR and the 95% confidence limit of the OR. In the panel i, a hyperbolic shape could not be fitted due to the considerable data variation, likely due to the inconsistency of the term “CVD” (see Supporting information: Method details). The numbers marked at each point are the same as the numbers shown in the original figure. The sources of each data are shown in Methods. This figure was re-used from Matsushita et al. Glob Heart. 2020; 15(1):64. Figure 5. https://www.ncbi.nlm.nih.gov/labs/pmc/articles/PMC7546112/figure/F5/ Copyright © 2020 The Authors. Creative Commons Attribution 4.0 International License (CC-BY 4.0) https://creativecommons.org/licenses/by/4.0/
Considering the cause of these patterns, we found the matching between grouping the data by hyperbolic patterns and grouping by data cut-off days (Fig. S11). We calculated weighted averages of age by grouping. In the earlier days group, non-severe was 46.5, and severe was 57.7, which were middle age (age 45-64). In the late date group, non-severe was 40.9, and severe was 68.6, which were mature age (age 25-44) and elderly (age > 65), respectively.
Surprisingly, we found unrecognized patterns during the data review process in a figure reported by Guo et al. (10), which showed a relationship between N-terminal pro-brain natriuretic peptide (NT-proBNP) and Troponin T (TnT). There were three clusters of subgroup patterns, and each cluster could be fitted parabola (Fig. 7 a, b, Fig. S12, a, b & Fig. S13, a, d, e-h).
a, Relationship between high sensitive C-reactive protein (hsCRP) and cardiac troponin T (TnT) in COVID-19 patients (left) and the relationship between cardiac troponin T (TnT) and N-terminal pro-brain natriuretic peptide (NT-proBNP) (right). b, Data points from the right side of panel a and fitting parabola. c, Three-dimensional data visualization was constructed by mounting the value of hsCRP onto panel b. d, Linear regression analysis on the side surface of the parabolic cylinder in panel c. The points were replaced with their average if the values could not be determined due to overlapping (the points indicated by the left-pointing arrow and the error bar, which are the average and the range of values, respectively). Panel a was re-used from Guo T et al. JAMA Cardiol. 2020; 5(7):811-818. Figure 1 https://www.ncbi.nlm.nih.gov/labs/pmc/articles/PMC7101506/figure/hoi200026f1/ Copyright © 2020 Guo T et al. JAMA Cardiology. Creative Commons Attribution License (CC-BY). https://creativecommons.org/licenses/by/4.0/
The third subgroup pattern was similar to ST-elevating myocardial infarction in the figure report by Budnik et al. (11). That study was a comparative study between Takotsubo cardiomyopathy (also known as stress cardiomyopathy or broken heart syndrome) and ST-elevating myocardial infarction. This tilted parabola appeared in another study on COVID-19 patients (Fig. S12, c) (43), and the pattern had already appeared in other studies (44, 45) before the pandemic (Fig. S13, b & c). However, patterns in COVID-19 patients were not recognized by each author.
Additionally, In the case of dropping points to the horizontal axis (troponin axis), the histogram on the troponin axis results in bimodality. The above results matched the patterns reported by other studies (Fig. S12, d & e) (43, 46, 47).
Furthermore, by mounting the data of high-sensitivity C-Reactive protein (hsCRP) onto the third subgroup in the data reported by Guo et al. (10) and constructing a parabolic cylinder, we found a pattern that might further be subdivided into two subgroups on the side surface of this parabolic cylinder (Fig. 7, c, d, & Fig. S14).
Discussion
Hyperbolic shape fitting in traveler’s thrombosis
In traveler’s thrombosis, we obtained consistent results from two analyses (Analysis 1A: 7.1 hours and 11.8 hours, Analysis 1B: 9.2 hours and 12.1 hours) (Fig. 3, b, Fig. S3, b). The discrepancy between the value of 7.1 hours and 9.2 hours seems to be explained by the miscalculation found in the data review process (see Table S1, Table S2, and Supporting information: Method details). Our results implied two S-curves. Also, we hypothesized two high-risk periods and two types of high-risk groups, which may have to be considered with the “factor V Leiden paradox” (48) (Supporting discussion 3: two high-risk periods and two types of high-risk groups?).
Cyclic patterns newly founded in this research and OC
Another noteworthy point is the cyclic patterns in the data reported by Cannegieter et al. (8) (Fig. 4). In Cannegieter et al. (8), the cyclic pattern disappeared around three months (90 days), consistent with the high-risk period in oral contraceptive (OC) use was the first three months (49). Since there is multi-phase OC in the category of OC, our findings imply that the 28 days cycle OC is attributable to the cyclic pattern.
Additionally, the law-risk period just before 90 days in the data reported by Kelman et al. (9) (Fig. 5). This law-risk period could be explained that extended-use type OC, which has a planned drug withdrawal, such as 84 active days and seven placebo days (50, 51).
The uptrend in the data by Kelman et al.(9) might be due to depot agent type OC administrated in 90 days cycle, which injects drugs periodically after decreasing the drugs in the body.
Based on the above results, types of OC use (multi-phase type, extended-use type, and depot agent type) might be triggered by travel.
The data reported by Cannegieter et al. contained many car trips (air travel, train, bus, and car trip). These cars might be honeymoon cars.
Kelman et al. described that although the risk increased within 2 to 4 weeks of flight (hazard period), the risk was low after the period because travelers were healthier than the average of a group (healthy traveler effect) (9). Also, Rothman mentioned the “persistence of travel-related risk” and “spline regression” (16). They might realize the position around 3-4 weeks as a valley between two distributions (honestly, we thought the same as them before we learned the cyclic patterns by spending long-time severe observations and non-linear regression analysis).
The difference in the cyclic patterns and HRT
Interestingly, there was a difference between the period of cyclic patterns (Cannegieter et al.: about 28 days, Kelman et al.: about 18 days). Also, the wave in the data reported by Cannegieter et al. was clear, but the wave in the data reported by Kelman et al. was not clear despite the exposures in Cannegieter et al. being more complex (air travel, train, bus, and car trip) (8) than Kelman et al. (only air travel) (9).
Regarding the halfway cycle (about 18 days) in Kelman et al. (9), it is notable that a combination of daily estrogen dosing and 10–14 days of progestogen (52) is used for HRT. Also, in 1999, Clérel and Caillard mentioned that OC and HRT might affect thrombosis after travel (34).
It is also notable that thrombosis onset in the data reported by Cannegieter et al. was “March 1999-March 2000” (only most late 1990s to early 2000) (8). Thrombosis onset in the data reported by Kelman et al. was “1981-1999” (including the whole of the 1990s) (9).
We found a slowdown of the increase of thrombosis onset at Paris airports in 1998, the year of HERS result was issued (Fig. S5). Also, our trend analysis showed that the number of literature on thrombosis decreased since around 2000 (Fig. S6). Moreover, our merged data from four studies (12, 33, 34, 39) on PE from the 1970s to the 2000s showed that the rate of women was most pronounced during the late 1990s (Table 1).
Therefore, the difference in waves can be explained as follows: the halfway cycle (17.9 days) in Kelman et al. (9) was a mixture of the 28 days OC cycle and a type of HRT cycle. However, only OC users might remain in the data reported by Cannegieter et al. (8) after the issue of the HERS study in 1998 (37).
Although the existence of the women patients not “menopausal” stage women (40s to 50s) but among around 70-75 years of age seems to disagree with HRT use (Fig. S4, d, & Fig. S8, d), the subjects of the WHI study were “postmenopausal” women (age 50-79) for prophylactic use for cardiovascular problems.
Men’s life stage and travel related phycological stress
Analysis 1B-related additional analysis suggested that the development of travel-related thrombosis occurs at specific ages (Fig. S7). On the other hand, Kelman et al. reported the incidence of VTE per 100,000 passengers (9). In this data (9), the incidence of VTE sharply increased after age 40, and it was a temporary decrease in the early 60s. Although it should be noted that these results are not limited to men, they are consistent with our results.
We also found that the relationship between male age and travel duration seems to be inverted-U-shaped (mountainous) (Fig. S9 & Fig. S10). Regarding this result, men’s life event-related psychological stress may be one of the risk factors.
Moreover, every four studies (12, 33, 34, 39) showed inconsistent results (correlated or inversely correlated). This situation seems to be the same as the parable of “blind men and an elephant.” Considering the correlation between age and the risk of thrombosis, the above phenomena may be one of the causes of controversial discussions.
Changing the framework of the disease concept of “travel-related thrombosis.”
There has been a widespread belief that in-flight environments cause thrombosis. Also, the phenomenon of the onset of thrombosis after over a few weeks to two months since the travel has been explained, such as the development of blood clots occurring on a cabin, but the dislodge of a thrombus occurs after (12). Also, although OC-related thrombosis had already been known, OC was recognized as not a substantial cause but one of the risk (aggravating) factors (13).
Other researchers might realize other possibilities changed the name of thrombosis after travel from “traveler’s thrombosis” to “travel-related thrombosis,” but they could not show evidence.
However, the above cyclic patterns of thrombosis onset may suggest that the disease concept of traveler’s thrombosis (travel-related thrombosis) should be changed. From the cyclic patterns, it is more reasonable to explain that the “travel” and “high-risk period of thrombosis with OC use (within three months from starting OC)” coincided with the travel (e.g., honeymoon and starting birth control OC).
On the other hand, although the results of a meta-regression analysis (Chandra et al.’s result (3) and our results of S-curve fittings) showing that the risk of thrombosis increases with travel duration have seemed to support the in-flight environment theory, the possibility may exist that what essentially correlates with the risk of thrombosis is “increased health risks with aging” or “psychological stress associated with rising job titles,” since men’s data shows an increase in travel time with age. Nemeth et al. experimentally showed that psychological stress (fear caused by a horror movie) was associated with a rise in blood coagulant factor VIII (53).
Recently, Er et al. reported a few cases of venous thromboembolism (VTE) incidence during the COVID-19 lockdown (54). However, all patients had inter mediate-high risk pulmonary embolisms. Therefore, the influence of the in-flight environment may be less than we had previously thought. From a comprehensive view, “traveler’s thrombosis” or “travel-related thrombosis” may require to be regarded as “life-event-related thrombosis.”
Even if a substantial risk factor is not the in-flight environment, it is not neglectable at the present research stage. Our thought that starting OC or HRT was triggered by travel suggests that usage based on risk diversification is important to avoid overlapping risks of drug and in-flight environment. Also, well-planned may allow more safe use of OC and HRT. Additionally, for men, perhaps we should focus more on “stress reduction.”
Application of our method for COVID-19 cardiovascular data
In applying our ideas for COVID-19, the dataset grouping by Matsushita et al.(41) matched the grouping by data cut-off dates (Fig. S11). Age structure diverged from middle age to mature and elderly. Middle-aged people (age 45-64) might enter the workforce for manual labor treating fish containers, mature people (age 25-44) might want to select IT jobs, and elderly people (age > 65) might stay in the house. So, our results were consistent with the hypothesis that COVID-19 spread from the seafood market.
In our data review process (Supporting discussion 4: linear regression analysis), our analysis of the data reported by Guo et al. (10) showed three clusters of subgroups and formed the tilted parabola. Also, the patterns matched the other studies (Fig. 7 a, b, Fig. S12 & Fig. S13) (11, 43–47). Considering the study by Budnik et al. (11), those subgroups may have different biological mechanisms. Also, we found patterns on the side surface of this parabolic cylinder (Fig. 7, c, d, & Fig. S14).
On the other hand, a paradoxical result on troponin was obtained just before the pandemic (55). Meisel et al. mentioned that the CRP to troponin ratio (CRP/troponin) could serve to differentiate between myopericarditis and acute myocardial ischemia (AMI), although not the study on COVID-19 patients. In COVID-19, Caro-Codón et al. reported interesting behavior of CRP (46). A meta-analysis by Lagunas-Rangel reported that the lymphocyte-to-C-reactive protein ratio (LCR) level, which was not a simple measurement value but a ratio, might be related to an inflammatory process (56).
The above studies might imply complex data structures in 2-dimensional and 3-dimensional scatter plots consisting of biomarker values, and our findings may serve the cardiac biomarkers’ research field.
Supplementary, our result suggested that the treatment of implicit function, which is unfamiliar in the medical and biological fields, is required. In this research, we used our method to fit the curves. However, a Fortran program created by Dr. Timmes (57), a researcher in astronomy, may also be helpful for future research.
Controversial discussions and overlooked patterns
From the consistent results of Analysis 1A and 1B and the explainable result of analysis 2, it is conceivable that our strategy and method were effective. Recently, we have thought that the concept of “medical research test” or “inspection of medical data analysis” can be developed analogically as an extension of the medical inspection, including laboratory tests for each patient for each patient. Like medical inspection using equipment, detecting overlayed hyperbolic shape patterns may be a tool or marker of some overlooked information (Fig. S1, e).
We could not judge whether our findings are the root cause of the remaining controversies in thrombosis research. However, judging from many findings, when conducting a literature review, although paying attention to the reproducibility of a pattern in charts is needed to avoid arbitrary discussions, more serious consideration should be given to a pattern on a chart that has been ignored as a noisy variation.
Conclusions
In the study of traveler’s thrombosis, there has long been a widespread belief that thrombosis is caused by the in-flight environment (e.g., immobility or dehydration). Also, the phenomenon of the onset of thrombosis after a long period since the travel, which seems to contradict the above idea, has been explained in a somewhat forced way, such as the formation of blood clots occurs on board the airplane but the dislodge of blood clots occur after the travel.
However, from our analysis results of the variation of bar charts which present the number of thrombosis cases, it is a more reasonable explanation that the travel and the initiation of oral contraceptive (OC) use were strongly linked so that the “travel” and “high-risk period of thrombosis with OC use” coincided with the travel. Also, although the results of a meta-regression analysis showing that the risk of thrombosis increases with travel duration have seemed to support the in-flight environment theory, the possibility may exist that what essentially correlates with the risk of thrombosis is aging or psychological stress associated with rising job titles, since men’s data shows an increase in travel time with age. Therefore, the influence of the in-flight environment may be less than we had previously thought.
Although simple linear regression analysis has been used in studies of cardiovascular biomarkers, including the case of COVID-19, based on the results of our observations by focusing on the reproducibility of patterns on scatterplots or by reconstructing 3D scatterplots from data on several panels, it seems that there is a complex scatterplot structure that does not suitable for simple linear fitting.
We could not judge whether our findings are the root cause of remaining controversial discussions in thrombosis research. However, regarding our findings, paying more attention to patterns in the chart is needed.
Data Availability
Data availability statement We analyzed clinical data published by other studies (third parties). Used data is identified by indicated information of citation (reference numbers and list of references). The corresponding author responds to inquiries in the case of measured values from published figures requested by reviewers or readers. Code availability statement Correspondence author (KK) can respond to inquiries for the corresponding author’s email address on offering the Python source code and spreadsheet software files for statistical analysis. Also, we have already uploaded our Python programs for the following GitHub repository to keep traceability perfectly. https://github.com/Keiichiro-KIMOTO-Kagoshima-University/Findings_in_Travel-related_Thrombosis_and_COVID-19_Related_Cardiac_Biomarker_Data
Statement of our intention for data analysis results on previous published articles
In addition to general acknowledgments, to clarify our stance, we mention this statement of intention for results. In this article, we pointed out many overlooked information and errors. However, we have no intention to attack previous works because our analysis results owing to their original works, including original research articles, reported valuable data, and writings of meta-analysis synthesized valuable datasets. Since just a scientist had better reconfirm the previous studies with no preconceptions and being grateful to those studies’ researchers, we carefully reviewed the data reported by previous studies. We highly respect previous works intending to solve medical issues.
Transparency declaration
This study was not required ethical approval because the study used ONLY two types of openly available human data. Firstly, this study performed meta-analyses as secondary analyses. The source of the data was described in the method section. Second, this study analyzed measured values from figures in previous published scientific papers.
All the figures were appropriately treated with careful attention to the copyright issues to protect the integrity of the science. The sources of the original articles were described in the figure legend following the copyright holder’s instructions. Also, we submit copyright details in another file to a journal that we submit this manuscript file to explain additional information.
This study suggests a new review strategy and meta-analysis method for exploratory data analysis and reports some epidemiological findings. So, there were no applicable guidelines (e.g., Preferred Reporting Items for Systematic reviews and Meta-Analyses, PRISMA). However, this study carefully follows the usage of the statistical analysis method, and the whole process is precisely described in the method details section and Fig. 2.
This study was not received financial support from any funding.
Data availability statement
We analyzed clinical data published by other studies (third parties). Used data is identified by indicated information of citation (reference numbers and list of references). The corresponding author responds to inquiries in the case of measured values from published figures requested by reviewers or readers.
Code availability statement
Correspondence author (KK) can respond to inquiries for the corresponding author’s email address on offering the Python source code and spreadsheet software files for statistical analysis. Also, we have already uploaded our Python programs for the following GitHub repository to keep traceability perfectly.
Acknowledgments
One of the authors, Keiichiro Kimoto, appreciates the kindful encouragement of Hideo Yoshioka, MEcon., who is in charge of the Data Strategy Research Institute representative.
Footnotes
Copyright
The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, an exclusive licence (or non exclusive for government employees) on a worldwide basis to the BMJ Publishing Group Ltd to permit this article (if accepted) to be published in BMJ editions and any other BMJPGL products and sublicences such use and exploit all subsidiary rights, as set out in our licence.
Competing interest statement
Munekazu Yamakuchi, Kazunori Takenouchi, and Teruto Hashiguchi have completed the Unified Competing Interest form and declare: no support from any organization for the submitted work; no financial relationships with any organizations that might have an interest in the submitted work in the previous three years, no other relationships or activities that could appear to have influenced the submitted work. Keiichiro Kimoto has been in charge of external advisor for the Data Strategy Research Institute but received no financial support for this study.
Classification: Research article
This file includes Abstract, Summary boxes, Main Text, Figures 1 to 7, and Table 1. Also, this article has supporting information that contains supporting figures, tables, method details, and supporting discussions.
We thought submitting our research to a general scientific journal would be appropriate because this manuscript has both importance in medicine and scientific importance. However, we considered comments that we should submit our manuscript to a journal specializing in medicine, and we decided the manuscript to such a journal. So, we revised our manuscript for readers in medical journals, especially the introduction and discussion. Also, we add a conclusion section.
References
