Access the full text.
Sign up today, get DeepDyve free for 14 days.
Meta-analyses suggest that the published literature represents only a small minority of the total data collected in biomedical research, with most becoming ‘dark data’ unreported in the literature. Dark data is due to publication bias toward novel results that confirm investigator hypotheses and omission of data that do not. Publication bias contributes to scientific irreproducibility and failures in bench-to-bedside translation. Sharing dark data by making it Findable, Accessible, Interoperable, and Reusable (FAIR) may reduce the burden of irreproducible science by increasing transparency and support data-driven discoveries beyond the lifecycle of the original study. We illustrate feasibility of dark data sharing by recovering original raw data from the Multicenter Animal Spinal Cord Injury Study (MASCIS), an NIH-funded multi-site preclinical drug trial conducted in the 1990s that tested efficacy of several therapies after a spinal cord injury (SCI). The original drug treatments did not produce clear positive results and MASCIS data were stored in boxes for more than two decades. The goal of the present study was to independently confirm published machine learning findings that perioperative blood pressure is a major predictor of SCI neuromotor outcome (Nielson et al., 2015). We recovered, digitized, and curated the data from 1125 rats from MASCIS. Analyses indicated that high perioperative blood pressure at the time of SCI is associated with poorer health and worse neuromotor outcomes in more severe SCI, whereas low perioperative blood pressure is associated with poorer health and worse neuromotor outcome in moderate SCI. These findings confirm and expand prior results that a narrow window of blood-pressure control optimizes outcome, and demonstrate the value of recovering dark data for assessing reproducibility of findings with implications for precision therapeutic approaches. . . . . . . . Keywords Data science Metascience Neurotrauma Reproducibility Spinal contusion Motor recovery Autonomic Hemodynamics Introduction the practice of science have identified shortcomings in schol- arly communications that limit the full potential of biomedical The current system of biomedical research has generated enor- research. Estimates suggest that only 50% of completed clin- mous gains in knowledge, helping improve health outcomes ical and preclinical studies are reported in the published liter- over the past century. However, meta-analyses focusing on ature (Chan et al., 2014). In addition, up to 85% of all * Jessica L. Nielson Department of Neurology, University of Texas, Austin, TX, USA email@example.com Department of Psychology, University of Texas, Austin, TX, USA * Adam R. Ferguson Department of Psychiatry and Behavioral Sciences, University of firstname.lastname@example.org Minnesota, Minneapolis, MN, USA Institute for Health Informatics, University of Minnesota, Department of Neurological Surgery, Weill Institute for Minneapolis, MN, USA Neurosciences, Brain and Spinal Injury Center, University of California San Francisco, San Francisco, CA, USA San Francisco Veterans Affairs Health Care System, San Francisco, CA, USA W.M. Keck Center for Collaborative Neuroscience, Rutgers University, New Brunswick, NJ, USA 40 Neuroinform (2022) 20:39–52 biomedical research investment in data collection fails to yield The goals of the present project were to recover these dark publications, equating to a loss of over $200 billion in research data and make them FAIR, and to perform a multicenter investment worldwide per year (Chalmers & Glasziou, 2009; replication/cross-validation of the previous single-center, Røttingen et al., 2013). A consequence of failure to publish is machine-learning discovery that MAP predicted neuromotor “dark data”, where large quantities of research data remain outcome (Nielson et al., 2015). MASCIS data from the Ohio locked away in hard-drives and file cabinets in formats diffi- State University was previously used as our hypothesis gen- cult to access by the public or other interested parties (CMAJ, eration dataset, where the finding about the negative impact of 2014). Furthermore, the published literature often reflects perioperative hypertension on SCI outcomes was discovered summaries of methods, protocols, and experimental results using a novel form of machine intelligence called topological (e.g., p values, means, standard errors, graphs), which are data analysis (TDA) (Nielson et al., 2015). In the present study not as informative as granular subject-level data used to derive we used data from the remaining 7 sites as external cross- these statistics (Chan et al., 2014). Making dark data accessi- validation data to test the reproducibility of this hypothesis ble would improve the return on research investment by using traditional, confirmatory analytics. granting more people access to re-analyze and explore scien- Our team worked with original MASCIS consortium mem- tific data (Ferguson et al., 2014). bers to recover additional multicenter preclinical data collect- To improve value of biomedical research investment, ed across the study sites. After assembling a larger, and more Mueck (2013) and Wilkinson, et al. (2016)proposedmaking representative MASCIS dataset from recovered paper records, raw biomedical research data Findable, Accessible, we tested whether the Nielson et al. (2015) finding could be Interoperable, and Reusable (FAIR). The FAIR data steward- independently replicated using recovered data from the other ship principles have been endorsed by the US National MASCIS sites. Concurrent with this publication, we are re- Institutes of Health (NIH) and major publishers. A major source leasing the recovered MASCIS data as a citable dataset of dark data are small granular data sets collected by laborato- (doi:https://doi.org/10.34945/F5QG66) through the newly ries over the course of day-to-day research, so called “long-tail formed Open Data Commons for SCI (http://ODC-SCI.org), data” (Ferguson et al., 2014). Long-tail data contain useful in- a public data sharing infrastructure (Callahan et al., 2017; formation such as non-targeted endpoints of experiments, alter- Fouad et al., 2020). This serves our two adjacent purposes: native measures, and pilot data. In addition, long-tail data in- providing meaningful scientific contributions to the field by clude results from failed experiments and ancillary records to cross-validating a clinically relevant finding, and converting published studies that were never published or disseminated. In MASCIS dark data and the millions of dollars spent on their animal research, such dark data often are recorded in veterinary acquisition (NIH R01 NS032000) into FAIR data that can care logs that are not considered primary endpoints in biomed- continue to fuel new discoveries for SCI research into the future (Wilkinson et al., 2016). ical experiments. Recent efforts to collect and analyze dark data using advanced machine learning have yielded new findings with clinical implications (Hawryluk et al., 2015; Hawryluk et al., 2020; Nielson et al., 2015; Readdy et al., 2016). Methods Specifically, by applying modern machine intelligence tools to archived data we discovered that mean arterial blood pressure MASCIS Data Between 1993 and 1997, the NIH funded a (MAP) in the perioperative phase of SCI is a robust predictor of consortium of eight laboratories (Wise Young, contact PI) to neuromotor recovery (Nielson et al., 2015). validate and standardize the MASCIS/NYU Impactor device This initial MAP finding relied on data recovered from one used to give rats contusive SCI (Constantini & Young, 1994), center from the multicenter animal spinal cord injury study and test promising treatments in a rat model for thoracic SCI. (MASCIS), a preclinical drug trial conducted in the 1990s to There were three studies in MASCIS. In 1994, the MP94 compliment the National Acute SCI Study (NASCIS) human study assessed the effects of methylprednisolone (MP) on clinical trials comparing several experimental therapies graded rat SCI across 3 injury severities (12.5, 25 and against the anti-inflammatory glucocorticoid methylpredniso- 50 mm weight drop contusions). The second study, in 1995 lone in thoracic SCI. MASCIS had an enormous impact on the (YM95), assessed the effects of thyrotropin releasing hormone spinal cord injury (SCI) field. The consortium developed and analogue YM14673 on the same SCI models. Both MP and validated the NYU-Impactor device to model contusive SCI YM14673 had been shown to improve recovery following a (Constantini & Young, 1994), and standardized a locomotor SCI (Behrmann et al., 1994; Constantini & Young, 1994; outcome scale for rats (Basso et al., 1995; Basso et al., 1996). Faden, 1989). The third study (MY96) compared MP94 and Both the NYU-Impactor and BBB locomotor scale remain YM95 protocols from the preceding years that the consortium widely used throughout preclinical SCI research (Young, determined were most successful. All centers executed the 2002). However, the results of the treatment effects in same methods and protocols, and all was approved by each MASCIS were never published. institution’s respective Institutional Animal Care and Use Neuroinform (2022) 20:39–52 41 Committee. The centers were: Ohio State University (Center archeologists gained access to the storage unit to search for 1), University of California - San Francisco (Center 3), Alfred MASCIS data from all study sites over two afternoons. We I. DuPont Institute, Georgetown University Medical Center, retrieved a few floppy disks that putatively contain records Medical University of South Carolina, New York University, from the study, but we only had partial success in retrieving University of Florida - Gainesville, and Washington these data due to a combination of format obsolescence and University School of Medicine (note we were not given ex- ‘bit rot’ that occurs as magnetic media ages. In addition, we plicit permission to re-identify these centers, but this informa- retrieved thousands of paper records which were scanned and tion is available upon request). The protocols established for converted to PDFs (Neff, 2018). The paper records were man- MP94 remained relatively unchanged until MY96. A notable ually curated and organized into spreadsheet files, as had been exception was the exclusion of the 50 mm injury severity, done with the first iteration of Ohio State University data which was too severe to reliably measure recovery. Because curated in the VISION-SCI repository (Nielson et al., 2014). we were not able to recover data enough data from YM95, our Based on the results reported in the current paper (Fig. 1)we current study focused on the data collected in MP94 and can surmise that the storage unit contains additional records MY96. buried within it or that data are lost to bit rot given the dis- crepancies in the intended sample size and the recovered sam- MASCIS Animals Briefly, adult rats (age 77 ± 2 days) were ple size. That said, we have little reason to presume that the randomly assigned to a graded contusion severity condition recovered data are not a representative sample of the popula- of either a 12.5, 25, or 50 mm weight drop for MP94, and only tion level effects. 12.5 and 25 mm for MY96 at thoracic level 9–10 (T9–10). Animals were assigned at random to a treatment group Data Entry and Curation After studying original documents, (MP94, Supplemental Table 1; MY96, Supplemental protocols, and data sheets, we created a digital data template to Table 2). All groups included equal number of males and digitize the data we recovered. When appropriate, we matched females, and animal assigned to different contusion severity data fields from the hard copy data sheets to common data conditions. Perioperative systolic and diastolic blood pressure elements (CDEs) used by SCI data repositories (Nielson was monitored after the animal was anesthetized during the et al., 2014). Unique data fields were created for variables that contusion surgery using an arterial catheter. The perioperative were not CDEs. blood pressure values were recorded three different times dur- As the data was digitized and the dataset populated, the ing the procedure: within 20 min before to the moment of the goal was to enter the data as it was originally collected and SCI; at the moment of injury which was distinguished by a written. However, some curation took place during data entry sharp spike in the blood pressure recording; and within the by fixing simple errors made by the original data creators. This level of curation required little field expertise, and in- 20 min after the injury. Rats assigned to the acute survival condition were euthanized 48 h post SCI, and those in the volved transforming data to its intended form. For example, chronic survival condition were evaluated using the Basso- if the original protocol requested an animal’s temperature to Beattie-Bresnahan (BBB) locomotor scale (Basso et al., be written in Celsius and the original data creator wrote the 1995; Basso et al., 1996) 2 days post SCI, and once per week temperature in Fahrenheit, we converted those value back to for 6 weeks. All data collection were performed under insti- Celsius. Another example was transforming values from mil- tutionally approved animal care and use committee protocols ligrams (mg) to micrograms (μg) when the protocol and data at the constituent sites, adhering to federal standards. sheet were intended for the data to be entered as micrograms. We also correct grammatical and spelling errors made by the Legacy Data Retrieval Our team worked with the original original data creator when appropriate (e.g., then vs. than MASCIS team to track down the multicenter preclinical data grammatical errors). collected by MASCIS across all sites. We learned that during Though rarely employed, occasionally we omitted data if it the trials, copies of all data sheets from all centers, including could not be accurately recovered, and this required some surgery records, outcome measures, notes, etc., were sent to domain expertise. Some information was lost or became illeg- NYU (the primary center for MASCIS) to be analyzed. The ible over the years, or when records were scanned for digiti- treatment protocols in the study did not return significant find- zation. In these cases, it was inappropriate to guess the original ings, and the results of the treatment effects were never pub- data, and our team opted to leave those data points empty and lished. Shortly after MASCIS was concluded in 1997, the PI considered them ‘missing’ in subsequent analysis. For exam- moved from New York University to Rutgers University and ple, if an animal’s weekly BBB scores read 2, 9, 11, 10, 12, 3, all data sheets, hard disks, computers, reports, protocols, and 11, the score second to last score was unexpected. While it is study materials were boxed up into 3 full sized moving trucks possible the second to last score was 3, it was also possible and stored in a commercial storage unit in Piscataway, New information was lost. It was inappropriate to fill in the expect- Jersey in a large storage unit (Neff, 2018). Our team of data ed value or guess the original score, but also inappropriate to 42 Neuroinform (2022) 20:39–52 Fig. 1 This flow chart describes the number of rats used in each of the paper describes data recovered from Centers 2–8, collectively titled three MASCIS studies, the number of rats for which data was recovered MASCIS 2020. Note that Center 2 did not contribute data in MY96, from each MASCIS study represented by solid or dashed lines, and where and Center 8 only contributed data in MY96. The compiled dataset can that data can be retrieved. An unknown number of rats were used in be retrieved in odc-sci.org and is titled “ODC-SCI MASCIS”. Upon YM95. Data from Center 1 (OSU) were used in Beattie et al., 1997, request or with permission, some centers can be unmasked Young 2002; Ferguson et al., 2004, and Nielson et al., 2015. The current disregard the legible value because it could be an outlier. In retrospective data from animal models of SCI (Ferguson et al., these situations we opted to consider the data missing in at- 2011, 2013;Nielson etal., 2014). VISION-SCI retrieved tempt to maintain data integrity and harnessed formal missing subject-level data of approximatively 3000 mice, rats, and values analysis and robust methods to make statistical infer- monkeys from 13 different laboratories from studies unpub- ences in the face of missingness (see statistical analysis). lished and published between 1993 and 2013. Part of the data After we digitized all of paper records, we began our post incorporated into VISION-SCI came the Ohio State site in the data entry curation by looking at the mean, median, mode, Multicenter Animal Spinal Cord Injury Study (MASCIS) and minimum, and max values to identify errors in our database. was reported in Nielson et al., 2014.For the purposesofthe For example, if the average temperature for a data field was current paper, these prior data from Nielson et al., 2014–2015 36.2 degrees Celsius and the mode is 35.9 degrees Celsius, but were excluded from analysis to reflect an independent repli- the max is value was 381 and the minimum was 3.72, the cation of the results with subjects from non-OSU sites. curator could fix those mistakes by assuming those were mis- takes made during the digitizing phase when the dataset was Statistical Analyses Our analysis was performed on the dataset being populated. When dealing with variables that required we created from the paper records recovered. After digitizing field expertise or when the curator was unsure if there was a and curating the records, data was analyzed using SPSS v25 mistake (e.g., anesthesia drug dose), the curator checked the (IBM Chicago, IL) and the statistical programing language R values on the original data record. When that was not an op- v3.6.0 (R Foundation for Statistical Computing, Vienna, tion, or when our team was unsure about the quality of the data Austria) with R Studio integrated development environment point after checking original records, we opted to leave the (R Core Team, 2019; RStudio Team, 2018). Missing values data point empty and consider the data point missing. analysis was run using SPSS, and the null hypothesis that values were missing completely at random (MCAR) was test- VISION-SCI Data Previously attempts to recover MASCIS data ed using Little’sMCAR test. were included in the Visualized Syndromic Information and Our MASCIS dataset included weekly values for BBB and Outcomes for Neurotrauma-SCI (VISION-SCI) database weight. Some animals had multiple scores per week, and an funded by the National Institute of Neurological Disorders aggregated score was calculated in those cases. We were in- terested in the effects of time measured in days post SCI, and Stroke (NINDS) to create a data repository by collecting Neuroinform (2022) 20:39–52 43 perioperative blood pressure, sex, and contusion drop height, validate treatment protocols, anesthesia, and outcome mea- on BBB locomotor recovery and weight gain. Using systolic sures (Supplemental Table 1)(Fig. 1). We recovered records and diastolic blood pressure values collected during SCI sur- from n = 252 rats with 2 days post SCI survival (Acute proto- gery, we estimated the perioperative mean arterial blood pres- col), and n = 489 rats with 6 weeks post SCI survival (Chronic sure values (MAP = (SBP + 2*DBP) / 3). Weight gain was protocol). Records were recovered for an additional n =31 calculated as the percent of change in weight (Δ%weight) animals, but we were unable to determine with certainty the from baseline weight (e.g., a 30 g weight gain by a 300 g intended survival time. In sum, we recovered records for 772 animal is 10% gain; Δ%weight = [weight / baseline weight] rats from MP94. The MASCIS 1996 protocol designated n = – 1 * 100). Locomotor recovery was the change in BBB score 504 rats for inclusion, and we recovered data for n =353 of (ΔBBB) calculated by subtracting the subject’s initial BBB them (Supplemental Table 2). Prior work reported on 132 rats score, recorded in the first 3 days post SCI, from the final from MP94 and 72 from MY96, from the OSU cohort (Center BBB score before the subject expired or was perfused. A 1) was described and accounted for in Nielson et al. (2015). previous study has demonstrated BBB score does not signif- We have excluded these data from analyses in the current icantly improve 22 days post SCI (Hook et al., 2004), so the work, but are making these dark data FAIR and releasing them final BBB score for a subject was used to calculate the change as a companion to the current paper (doi:https://doi.org/10. in BBB as long as locomotor evaluation took place 25 days 34945/F5QG66). Assuming all planned animals were post SCI. included in the experiments and excluding animals from We tested 4 separate Linear Mixed Models (LMM). OSU, our recovery rate for MASCIS 1994 was 72.28% of Models were generated using lmer function in the R package animals, and 81.71% of animals for MASCIS 1996 (Fig. 1). lme4 (Bates et al., 2014), and the lmerTest package generated There are 296 rats unaccounted for from MP94, 79 from the Type III Analysis of Variance Table with Satterthwaite’s MY96, and hundreds from YM95. During the data archeology method (Kuznetsova et al., 2017). In the first pair of LMMs, expedition, we were not able to search all boxes in the storage we assessed BBB locomotor recovery after SCI as the out- unit, and the unaccounted for records may still be in storage. come variable. Pre-injury blood pressure (MAP) collected Moreover, some centers may not have sent copies of all their within 20 min of SCI was a fixed factor the first LMM, and records to NYU before the study ended (e.g., in MASCIS blood pressure (MAP) at time of injury (distinguished by a 1996 internal progress reports, Center 4 contributed 11 of 72 sharp spike in the blood pressure recording) was a fixed factor records from the planned subject count, and Center 8 contrib- in the second LMM. In addition to pre-injury or at-injury uted 0 of 72 planned subject counts), which may partially blood pressure, time (days post SCI) and contusion severity explain why the number of animals per center in our dataset (drop height) were fixed factors. Center and subject with a is not evenly distributed. random slope by time were the random factor. For the third Of the 1125 rats in our recovered dataset from MASCIS and fourth LMMs, we assessed weight gain after SCI as the 1994 and MASCIS 1996, none have data records that were outcome variable, as weight after injury is frequently used as a complete both within a test date and across all possible repeat- general measure of health and wellbeing. Pre-injury blood ed measures (time-points). It is not always clear when post- pressure (MAP) was a fixed factor the third LMM, and at- operative records ended because perfusion dates were not al- injury blood pressure (MAP) was a fixed factor in the fourth ways recorded in the perfusion logs we scanned. However, we LMM. In addition to pre-injury or at-injury blood pressure, were able to estimate our overall data recovery rate based on time (days post SCI) and contusion severity (drop height) surgical records, which describes the subject’s surgery and were fixed factors. Center, sex, and subject with a random condition for the first 48 h post injury. Surgery record sheets slope by time were the random factor. To explore interactions had 64 primary variables (Supplemental Table 3). Of the 1125 effects from the LMMs, we applied general linear models rats, n = 1121 had surgery records with at least 1 of the 64 (GLM) using the lm function in R. Eta squared values were variables completed. Our overall data recovery rate for sur- generated using the sjstats package in R (Lüdecke, 2020)or gery records was 60.44% (Fig. 2). This value might under- computed by custom code. estimate recoverable data. Some rats died within 48 h of inju- ry, while others were excluded from the study for reasons noted in their surgery sheets including anesthesia dosage, or Results surgery complications. For these reasons, we suspect portions of some surgery records were blank on purpose. We are con- Data Provenance and Descriptive Statistics on fident that n = 500 survived postoperative complications be- Recovered Data cause we recovered at least one data point collected at least 48 h post SCI from perfusion or post-operative care records. According to the original MASCIS protocol, 1200 rats were For those n = 500 rats, our surgery related data recovery rate was slightly better at 64.63%. Our objective is not to revisit planned for inclusion in 1994, including experiments to 44 Neuroinform (2022) 20:39–52 data collection practices, nor compare data collection between analysis while mitigating missing values (Nielson et al., centers. However, a data-driven missing values analysis dem- 2020). onstrated that values were not missing completely at random The first LMM targeted pre-injury MAP as a predictor of (Little’sMCARtest, p <0.01). BBB locomotor scores using 2327 observations from 441 Although we recovered substantial information from unique rats across 6 Centers in the (Table 1; Fig. 3a). We MASCIS, including original protocol and internal progress found significant main effects for contusion severity and time reports, we were not able to decipher the drug treatment on BBB scores. Animals with more severe contusions had blinding codes, and thus we cannot report on the results of worse BBB scores, and BBB scores improved as recovery the drug treatment protocol at the current time. Treatments time increased. There was also a significant three-way inter- were spread evenly between centers, animal sex, and contu- action between the pre-injury MAP, time post SCI, and con- sion severity. According to the MASCIS progress reports, tusion severity on BBB scores. This indicates that pre-injury none of the MASCIS 1994 treatments resulted in better out- blood pressure correlated with recovery of function, but this comes compared to control, and some treatments may have effect had different directionality depending on injury resulted in worse recovery and possibly death. The methyl- severity. prednisolone 1 treatment in MASCIS 1996 resulted in better In the second LMM targeted at-injury MAP as a predictor BBB recovery compared to saline. Moreover, according to of BBB locomotor outcome scores using 1081 observations unpublished MASCIS progress reports submitted to NIH, among 197 unique rats across 4 Centers (Table 2). There were the independent variable that had the largest effect size on significant main effect for contusion severity and time on outcome was center. This opens the possibility that nuisance BBB scores. Locomotor scores improved as recovery time variables associated with specific centers drive the majority of increased, and scores decreased as contusion severity in- the variance in outcome, potentially occluding drug effects. creased. There was a significant two-way interaction between Our prior work strongly suggests that uncontrolled variance in at-injury MAP and contusion severity, and a three-way inter- operative blood pressure may be one such variable (Nielson action between at-injury MAP, time post SCI, and contusion et al., 2015). severity on BBB scores. Post hoc analyses were required to understand the precise Confirmatory Hypothesis Testing of the Blood nature of the significant blood pressure-recovery interactions Pressure-Locomotor Recovery Association uncovered by LMM analyses. We used a GLM to assess the effect of pre-injury MAP and contusion severity on ΔBBB We used LMMs to test the relationship between perioperative from baseline to the time of the rat’s expiration (Table 3). (20 min pre-injury and at-injury) blood pressure and BBB As showninFig. 3b, for moderate SCI (12.5, 25 mm recovery, marshaling all available data recovered for each weight-drop), higher pre-injury MAP associated with better Fig. 2 This heatmap demonstrates data recovered from the surgery record sheets of MP94 and MY96. Each row represents a unique rat (n = 1125), and each column represents a unique variable from the surgery records (n = 64), and each individual square a data point Neuroinform (2022) 20:39–52 45 Table 1 Linear Mixed Model Output of BBB with Pre-Injury Blood Pressure as Fixed Factor Variables NumDf DenDf F-Value p value η Pre-Injury Blood Pressure 1 506.38 1.5174 0.2186 0.002 Contusion Severity 2 435.62 8.6503 0.0002 0.038 Days Post SCI 1 788.47 147.1760 <0.0001 0.157 Pre-Injury Blood Pressure x Contusion Severity 2 430.87 0.0242 0.9761 <0.001 Pre-Injury Blood Pressure x Days Post SCI 1 771.89 0.5725 0.4495 <0.001 Pre-Injury Blood Pressure x Contusion Severity x Days Post SCI 2 743.54 9.7646 <0.0001 0.025 outcome, whereas for severe SCI (50 mm) higher MAP asso- significant two-way interactions between pre-injury MAP ciated with worse outcomes. This indicates that pre-injury and time and pre-injury MAP and contusion severity on blood pressure affected recovery of function, and this relation- Δ%weight. Lastly, we found a significant three-way interac- ship had different directionality depending on injury severity, tion between contusion severity, time post injury, and pre- with more severe SCI demonstrating a more profound nega- injury MAP on Δ%weight gain. tive influence of high blood pressure. The same form of effect The fourth LMM included 1349 unique observations be- was observed with at-injury MAP, suggesting that periopera- tween 276 rats across 3 Centers (Table 5). In this model, we tive blood pressure is a robust predictor of BBB. found a significant main effect of time on Δ%weight, and a The third LMM targeted the pre-injury MAP on weight significant three-way interaction between at-injury MAP, con- gain using 2336 unique observations among 414 rats across tusion severity, and time on Δ%weight. 5 Centers (Table 4; Fig. 3c). There were significant main Post hoc GLM was required to further understand the ef- effect for pre-injury MAP and time on Δ%weight, and fects on Δ%weight. This analysis revealed a marginally Fig. 3 Change in BBB score (a) and weight gain (c) over time are shown with SEM bars for each time point. The linear relations between pre-injury blood pressure and ΔBBB (b) and Δ%weight (d) depicted, and the shaded areas represents the 95% confidence interval 46 Neuroinform (2022) 20:39–52 Table 2 Linear Mixed Model Variables NumDf DenDf F-Value p value η Output of BBB with Blood Pressure at SCI as Fixed Factor At-Injury Blood Pressure 1 398.70 0.1556 0.6935 <0.001 Contusion Severity 2 208.55 8.9549 0.0002 0.079 Days Post SCI 1 119.22 39.2645 <0.0001 0.247 At-Injury Blood Pressure x Contusion Severity 2 220.07 3.9490 0.0207 0.034 At-Injury Blood Pressure x Days Post SCI 1 118.61 0.0972 0.7557 <0.001 At-Injury Blood Pressure x Contusion Severity x Days 2 125.52 9.8191 0.0001 0.135 Post SCI significant interaction effect between pre-injury MAP and had less recovery, while rats with milder injuries and higher contusion severity (Table 6). For animals in the low and me- blood pressure showed better recovery. To our knowledge, dium contusion severity conditions, higher blood pressure was this is the first time such an interaction between blood pres- associated with more weight recovery. The inverse was true sure, injury severity, and recovery has been demonstrated in for the high contusion severity group, where higher blood cases of SCI. In achieving our goal of cross validating a prior pressure was associated with less weight recovery (Fig. 3d). finding of high clinical import, we recovered value from the The analysis of weight change also showed that males recov- initial millions of dollars of investment by the NIH made over ered and gained more weight compared to females, but there two decades ago in the original MASCIS trials, and demon- were no significant interaction effects that included the sex of strated a practical application FAIR data principles (Table 7). the animal, suggesting the interaction between perioperative In addition, the results have direct implications for clinical blood pressure and contusion severity is not significantly dif- care in acute SCI. Managing MAP in acute SCI may be crit- ferent between males and females. ically important for preventing secondary injuries and neuro- We did not find meaningful results when we included post- logical deficits. In the published guidelines for acute medical injury MAP as fixed factors in our LMMs (results not shown), and surgical management of SCI, the American Association and this analysis is confounded by known effects of injury of Neurological Surgeons (AANS) and Congress of severity on subsequent autonomic derangements (i.e., poten- Neurological Surgeons (CNS) supported maintenance of tial for associations reflecting ‘reverse causality’ with SCI MAP above 85 and 90 mmHg for patients during the first severity) (Nout et al., 2012). Altogether, the results suggest week after admission (Hadley et al., 2002; Walters et al., that perioperative hypertension is associated with poorer 2013; Yue et al., 2017). The rationale is that low blood pres- health and worse locomotor recovery in more severe SCI sure reduces blood flow and patients that are kept at a higher whereas perioperative hypotension is associated with poorer MAP after SCI show better recovery (Casha & Christie, 2011; health and worse recovery in moderate SCI. Catapano et al., 2016;Dakson et al., 2017; Hawryluk et al., 2015; Sabit et al., 2018). Nielson et al. (2015) were the first to note that hypertension, in addition to hypotension, impairs recovery. Discussion One of the important implications of our findings pertains to precision medicine. Kepler et al. (2015)reported thatpa- In the current confirmatory study, we recovered legacy data from 1125 rats to independently replicate the results from tients with pre-existing hypertension had worse recovery com- pared to controls. They proposed that blood pressure goals for Nielson et al. (2015). Our results suggest an interaction effect between perioperative blood pressure and contusion severity, those patients may have to be set even higher than those rec- ommended by AANS and CNS, and further studies are where rats with more severe injuries and higher blood pressure Table 3 General Linear Model Variables Df F-Value p value η Comparing ΔBBB Across Groups Pre-Injury Blood Pressure 1 1.4429 0.2309 0.0050 Contusion Severity 2 19.6133 <0.0001 0.1420 Pre-Injury Blood Pressure x Contusion Severity 2 3.1911 0.0430 0.0230 Residual 230 Neuroinform (2022) 20:39–52 47 Table 4 Linear Mixed Model Variables NumDf DenDf F-Value p value η Output of Weight Gain with Pre- Injury Blood Pressure as Fixed Pre-Injury Blood Pressure 1 393.72 5.4686 0.0199 0.013 Factor Contusion Severity 2 406.60 2.0256 0.1332 0.009 Days Post SCI 1 272.51 8.8732 0.0032 0.031 Pre-Injury Blood Pressure x Contusion Severity 2 405.77 2.9243 0.0548 0.014 Pre-Injury Blood Pressure x Days Post SCI 1 268.60 21.4986 <0.0001 0.074 Pre-Injury Blood Pressure x Contusion Severity x Days 2 263.72 11.6871 <0.0001 0.081 Post SCI needed to identify the role of hypertension, blood flow to the outcome. The first of these was recently published, in the form spinal cord, and recovery (Kepler et al., 2015). We agree that of a case series providing preliminary clinical support for the such studies are needed, due to the lack of consensus in clin- hypothesis (Ehsanian et al., 2020). Physiologically, it would ical protocol guidelines for maximum blood pressure for pa- stand to reason that hypertension may result in ‘hemorrhagic tients after SCI. The Center for Disease Control and conversion’, and exacerbate bleeding into the spinal cord and Prevention estimates one in three Americans are hypertensive resulting in tissue damage. In the animal literature it is well (Center for Disease Control and Prevention, 2020), and data is established that SCI compromises the blood-spinal-cord bar- needed to identify MAP goals that would maintain tissue rier and that peripheral blood components contribute to sec- function without impairing the neurological recovery for that ondary cell death, including infiltration of circulating immune population. Recent retrospective clinical studies of high- cells, circulating cytokines and other factors (Crowe et al., resolution physiological monitoring further supports MAP 1997;Ferguson etal., 2008;Kigerl etal., 2009). should be maintained above 85–90 mmHg up to seven days The major limitation in our analysis is that it is correlation- upon the patient’s admission to a hospital, and the proportion al, and not causal. In addition, our conclusions come from of time below 85 mmHg correlated with impaired recovery incomplete retrospective data. Not all of the original data (Hawryluk et al., 2015; Sabit et al., 2018; Walters et al., 2013). was recovered, and some may be permanently lost due format Physiologically, the rational is that spinal cord perfusion pres- obsolescence and bit rot of magnetic media. Not included in sure depends on systemic MAP remaining high enough to the data recovered were the drug treatment codes. The rats sustain tissue oxygenation in the injury penumbra in the face were treated with various drugs, and we remain blinded to of vertebral fracture and cord compression (Squair et al., 2019; their treatment condition. According to the MASCIS progress Yue et al., 2020). This SCI clinical guideline mirrors the logic reports, all but one treatment condition did not show a signif- of intracranial pressure monitoring in traumatic brain injury icant recovery associated with treatment. However, this does and other fields of cranial neurosurgery where prevention of not rule out the possibility that specific dose-response and hypotension using fluids and vasopressors is used to maintain timing features for methylprednisolone and other tested drugs intracranial pressure and decompressive hemicraniectomy is may have impacted the results. In addition, variation in animal used to prevent pressure overshoot (Chesnut et al., 2020; Shah care may also have introduced confounds. For example, et al., 2019). However, the concept of hypertension as a driver MASCIS used the anesthetic pentobarbital, which is known of poor outcome is less well established. In the wake of to produce blood pressure complications (Nout et al., 2012). Nielson et al., 2015 several clinical groups have begun explor- Some centers closely monitored blood oxygenation and per- ing hypertension as a potential negative prognosticator of formed resuscitation as needed, whereas other centers were Table 5 Linear Mixed Model Variables NumDf DenDf F-Value p value η Output of Weight Gain with Blood Pressure at SCI as Fixed At-Injury Blood Pressure 1 284.93 0.5197 0.4716 0.002 Factor Contusion Severity 2 255.31 1.1172 0.3288 0.008 Days Post SCI 1 118.83 24.0393 <0.0001 0.168 At-Injury Blood Pressure x Contusion Severity 2 248.99 1.0228 0.3611 0.008 At-Injury Blood Pressure x Days Post SCI 1 117.75 2.3158 0.1307 0.019 At-Injury Blood Pressure x Contusion Severity by Days 2 129.89 20.2185 <0.0001 0.237 Post SCI 48 Neuroinform (2022) 20:39–52 Table 6 General Linear Model Variables Df F-Value p value η Comparing Δ%weight Across Groups Pre-Injury Blood Pressure 1 10.0654 0.0017 0.025 Contusion Severity 2 16.5195 <0.0001 0.083 Sex 1 132.8158 <0.0001 0.335 Pre-Injury Blood Pressure x Contusion 2 2.7582 0.0657 0.014 Pre-Injury Blood Pressure x Sex 1 0.0989 0.7535 <0.001 Contusion x Sex 2 0.2040 0.8156 0.001 Pre-Injury Blood Pressure x Contusion x Sex 2 0.6046 0.5472 0.003 Residual 213 Table 7 FAIR data principles checklist for BPM replication in MASCIS Principle Definition Compliance F – FINDABLE To be Findable: Original MASCIS data F1. (meta)data are assigned a globally unique and persistent identifier F1 – yes F2. data are described with rich metadata (defined by R1 below) F2 – no F3. metadata clearly and explicitly include the identifier of the data it describes F3 – yes F4. (meta)data are registered or indexed in a searchable resource F4 – no MASCIS data entered into VISION-SCI/ODC-SCI F1 – yes F2 – yes F3 – yes F4 – yes A – ACCESSIBLE To be Accessible: Original MASCIS data A1. (meta)data are retrievable by their identifier using a standardized A1 – maybe communications protocol A1.1 – no A1.1 the protocol is open, free, and universally implementable A1.2 – N/A (unknown) A1.2 the protocol allows for an authentication and authorization A2 – maybe (in the protocols/grants?) procedure, where necessary MASCIS data entered into A2. metadata are accessible, even when the data are no longer available VISION-SCI/ODC-SCI A1 – yes A1.1 – yes A1.2 – yes A2 – yes I – To be Interoperable: Original MASCIS data INTEROPERA- I1. (meta)data use a formal, accessible, shared, and broadly applicable language for I1 – no BLE knowledge representation. I2 – no I2. (meta)data use vocabularies that follow FAIR principles I3 – no I3. (meta)data include qualified references to other (meta)data MASCIS data entered into VISION-SCI/ODC-SCI I1 – yes I2 – yes I3 – yes R - REUSABLE To be Reusable: Original MASCIS data R1. meta(data) are richly described with a plurality of accurate and relevant attributes R1 – no R1.1. (meta)data are released with a clear and accessible data usage license R1.1 – no R1.2. (meta)data are associated with detailed provenance R1.2 – yes R1.3. (meta)data meet domain-relevant community standards R1.3 – yes MASCIS data entered into VISION-SCI/ODC-SCI R1 – yes R1.1 – yes R1.2 – yes R1.3 – yes Neuroinform (2022) 20:39–52 49 less focused on these anesthetic complications. In addition, biomedical research. Accelerating the transition from a univariate post-operative care protocols evolved over time, especially to a multivariate view of diseases should be a target for biomed- with respect to bladder care and antibiotic use to control mor- icine, and making data FAIR through data sharing and data ar- tality due to urinary tract infections. One of the centers dis- cheology are crucial and achievable steps in making that transi- covered the fluoroquinolone Baytril was highly effective at tion (Callahan et al., 2017; Ferguson et al., 2011, 2013;Fouad reducing post-SCI mortality, and this was later adopted by et al., 2020). the other centers. Accordingly, analyses by the original While there are reservations about data sharing among classi- MASCIS consortium determined the independent variable cally trained biology researchers, the –omics science disciplines that affected outcomes most was center, and data recovered have successfully navigated those concerns for decades (Kaye was not evenly distributed across centers. This suggests that et al., 2009; Lander, 1996). Genomic and other –omic published there is substantial variability in healthcare records, even in a studies always provide the accession number to National Center well-controlled and protocolized randomized control trial for Biotechnology Information (NCBI) datasets used for their (RCT) in animal subjects which have greater standardization analyses, where all datasets are publicly available for download. of housing, diet, health care and study conditions than a hu- In addition, many authors make their codes and scripts publicly man RCT. The fact that large center effects persist even under available on platforms like GitHub (GitHub Inc., San Francisco, these idealized conditions may be due to the fact that random- CA) for anybody to replicate and validate their analysis. While ization is performed and monitored in a small number of this may be a novel concept for some, members of neurotrauma indexed variables and may not apply to non-indexed variables disciplines have established pathways for data sharing (Fisher such as high blood pressure in MASCIS. Whether center-to- et al., 2009; Fouad et al., 2020;Huertaetal.,s 1993;Lemmon center variability is less in animals versus human RCTs, or et al., 2014;Marmarouetal., 2007). controlled trials versus observational is an interesting open Despite limitations, our study shows that even legacy data question that FAIR data sharing may help resolve in the fu- from 25 years ago may yield important findings, and this helps ture. Making individual participant data FAIR could enable support emerging standards that all NIH funded research should translational cross-walk meta-analysis between humans and follow FAIR data stewardship principles (Mueck, 2013; animals, if privacy and security concerns that arise from mul- Wilkinson et al., 2016). The first attempt to gather subject-level tidimensional clinical data can be appropriately mitigated data from neurotrauma studies was VISION-SCI (Nielson et al., (Rocher et al., 2019). Although we statistically controlled for 2014), but to our knowledge the present work represents the first the effect center in the present paper, and confirmed blood targeted attempt of data retrieval of animal subject level data at pressure effects, this post-hoc statistical approach is less pow- this scale. The MASCIS consortium was a large and expensive erful than a balanced prospective study for inferring causal group with a budget that exceeded $1 million annually between relationships. We therefore recommend a prospective study 1994 and 1996, and used over 2000 animals for their experi- assessing the impact of hypertension on recovery after SCI ments. Our inability to recover the original treatment conditions of different severities, where center and treatment differences for rats from MASCIS is not unusual given the regulatory stan- can be more directly controlled for. dards under which these data were collected. For the majority of Neurological trauma and related disorders are incredibly com- grant funded research, historically, NIH mandated that data be plicated to treat. Due to the complexity and heterogeneity of SCI maintained for 3–5 years post-study completion (NIH Office of and central nervous system (CNS) disorders, our viewpoint is Extramural Research, 2019). Having retrieved data for over 1000 that researchers would benefit by approaching these diseases as animals at an estimated data recovery rate above 60%, our expe- ‘big-data’ problem, specifically involving big data variety rience retrieving part of that dataset was overall successful be- (Ferguson et al., 2011; Hawkins et al., 2019;Huieetal., 2018). cause we increased the retained value from the original invest- SCI may result in motor control and mobility impairments; im- ment. Additionally, we are adding these data to our prior recov- paired breathing and respiratory deficits; loss of bladder function; ered data from OSU in our public release of the MASCIS data as bowel and sexual dysfunctions; pathological pain; and/or loss of part of this paper yielding a total of 1459 animals data records autonomy. To capture the multivariate syndromic outcomes of made FAIR through data archeology. CNS disorders, researchers often collect multiple outcome mea- While data archeology may increase the initial investment in sures for each individual subject. However, outcomes are often some circumstances, such as those presented here from the only assessed a few factors at a time. Complex and contemporary MASCIS study, we strongly recommend and endorse pursuing analytical methods, including those more easily associated with – a policy of applying FAIR data principles for neuroscience as omics, which permit researchers to explore the multi- data are collected, and specifically making raw subject level data dimensionality of diseases rather than testing a few factors at a accessible to the greater scientific community. Efforts to incor- time are becoming increasingly more accessible and common in porate this into study designs at the onset of data collection would biomedicine (Parikshak et al., 2015), and as was the case with ensure FAIR data access moving forward. While it is unclear Nielson et al. (2015) these methods will continue drive future whether data archeology is as laborious as prospective data 50 Neuroinform (2022) 20:39–52 Open Access This article is licensed under a Creative Commons collection, the scientific community risks losing data if is not Attribution 4.0 International License, which permits use, sharing, adap- collected and disseminated adherent to FAIR principles as we tation, distribution and reproduction in any medium or format, as long as demonstrated in this project. The NIH and NINDS CDEs greatly you give appropriate credit to the original author(s) and the source, pro- facilitate the opportunities of researchers sharing data among vide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included collaborators or colleagues, and new platforms to facilitating data in the article's Creative Commons licence, unless indicated otherwise in a sharing already exist or will soon be available for many disci- credit line to the material. If material is not included in the article's plines in biomedical research (Hawkins et al., 2019). Creative Commons licence and your intended use is not permitted by The present work extends the concept of meta-analysis to raw statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this source data, which opens new possibilities to develop higher licence, visit http://creativecommons.org/licenses/by/4.0/. evidence for preclinical studies (currently classed as level 4–5 evidence) (Biering-Sørensen, 2005). At the current time it re- References mains unclear to what extent systematic reviews and meta- analyses can be relied upon to be correct reflections of raw data Basso, D. M., Beattie, M. S., & Bresnahan, J. C. (1995). A sensitive and (Gøtzsche et al., 2007) as they are based primarily on statistics reliable Locomotor rating scale for open field testing in rats. Journal of reported in papers. In addition, reviews suggest that over 80% of Neurotrauma, 12(1), 1–21. https://doi.org/10.1089/neu.1995.12.1. published manuscripts in a biomedical science journal contains a Basso, D. M., Beattie, M. S., Bresnahan, J. C., Anderson, D. K., Faden, A. I., Gruner, J. A., Holford, T. R., Hsu, C. Y., Noble, L. J., Nockels, least one statistical error (Simundic & Nikolac, 2009), and there R., Perot, P. L., Salzman, S. K., & Young, W. (1996). MASCIS are no indications that statistical rigor is increasing in biomedical evaluation of open field Locomotor scores: Effects of experience research (Ercan, 2015; Ercan et al., 2017). Thus, there is much to and teamwork on reliability. Journal of Neurotrauma, 13(7), 343– be gained by granting the next generation of scientist’saccessto 359. https://doi.org/10.1089/neu.1996.13.343. FAIR datasets derived by data archeology, data recovery, and Bates, D., Mächler, M., Bolker, B., & Walker, S. (2014). Fitting linear mixed-effects models using lme4. ArXiv:1406.5823 [Stat]. http:// application of modern data stewardship and analytic tools of the arxiv.org/abs/1406.5823 sort applied here. Beattie, M. S., Bresnahan, J. C., Komon, J., Tovar, C. A., Hsu, C. Y., Noble, L. J., Salzman, S., & Young, W. (1997). Endogenous repair after spinal cord contusion injuries in the rat. Experimental Information Sharing Statement Neurology, 148,453–463. Behrmann, D., Bresnahan, J., & Beattie, M. (1994). Modeling of acute Our analysis code reproducing our figures and results are spinal cord injury in the rat: Neuroprotection and enhanced recovery available as part of the supplementary material online. All data with methylprednisolone, U-74006F and YM-14673. Experimental collated from MP94 and MY96 are available at: doi:https:// Neurology, 126(1), 61–75. Biering-Sørensen, F. (2005). Evidence-based medicine in treatment and doi.org/10.34945/F5QG66. rehabilitation of spinal cord injured. Spinal Cord, 43(10), 587–592. https://doi.org/10.1038/sj.sc.3101815. Supplementary Information The online version contains supplementary Callahan, A., Anderson, K. D., Beattie, M. S., Bixby, J. L., Ferguson, A. material available at https://doi.org/10.1007/s12021-021-09512-z. R., Fouad, K., Jakeman, L. B., Nielson, J. L., Popovich, P. G., Schwab, J. M., & Lemmon, V. P. (2017). Developing a data sharing Acknowledgments Supported by Wings for Life Foundation (ARF); community for spinal cord injury research. Experimental NIH/NINDS: R01NS088475 (ARF); UG3/UH3NS106899 (ARF); JLN Neurology, 295, 135–143. https://doi.org/10.1016/j.expneurol. supported by NIH/NIMH: R01MH116156 and NIH/NCATS: 2017.05.012. UL1TR002494; Department of Veterans Affairs: 1I01RX002245 Casha, S., & Christie, S. (2011). A systematic review of intensive cardiopul- (ARF), I01RX002787 (ARF); and Craig H. Neilsen Foundation (ARF). monary management after spinal cord injury. Journal of Neurotrauma, Original MASCIS data collection was funded by NS032000 (WY). The 28(8), 1479–1495. https://doi.org/10.1089/neu.2009.1156. authors would like to thank Hadi Askari, Sean P. O’Leary, and Patricia Catapano, J. S., Hawryluk, G. W. J., Whetstone, W., Saigal, R., Ferguson, Morton for help with data recovery. A., Talbott, J., Bresnahan, J., Dhall, S., Pan, J., Beattie, M., & Manley, G. (2016). Higher mean arterial pressure values correlate Author Contributions C.A.A. and J.L.N. drafted the manuscript. C.A.A., with neurologic improvement in patients with initially complete A.T.E., J.L.N., J.R.H. and A.R.F. performed analyses. J.L.N. and A.R.F. spinal cord injuries. World Neurosurgery, 96,72–79. https://doi. conceived the retrospective study. J.L.N., A.R.F., D.S., W.Y., M.S.B., org/10.1016/j.wneu.2016.08.053. and J.C.B. were involved in recovery, and interpretation of original data Center for Disease Control and Prevention. (2020). High blood pressure sets. C.A.A., J.L.N., and A.R.F. all contributed to data wrangling of fact sheet. Cdc.Gov. https://www.cdc.gov/bloodpressure/facts.htm? original data. W.Y., L.J.N., C.Y.H., J.C.B., and M.S.B. collected primary CDC_AA_refVal=https%3A%2F%2Fwww.cdc.gov%2Fdhdsp% experimental data. C.A.A., J.L.N., A.T.E., J.R.H., J.C.B., M.S.B., and 2Fdata_statistics%2Ffact_sheets%2Ffs_bloodpressure.htm A.R.F. contributed to the interpretation of the results. Chalmers, I., & Glasziou, P. (2009). Avoidable waste in the production and reporting of research evidence.,374,4. Declarations Chan, A.-W., Song, F., Vickers, A., Jefferson, T., Dickersin, K., Gøtzsche, P. C., Krumholz, H. M., Ghersi, D., & van der Worp, Competing Financial Interests All authors declare no competing finan- H. B. (2014). Increasing value and reducing waste: Addressing in- accessible research. The Lancet, 383(9913), 257–266. https://doi. cial interests. org/10.1016/S0140-6736(13)62296-5. Neuroinform (2022) 20:39–52 51 Chesnut, R., Aguilera, S., Buki, A., Bulger, E., Citerio, G., Cooper, D. J., roundtable preclinical recommendations. Stroke, 40(6), 2244–2250. https://doi.org/10.1161/STROKEAHA.108.541128. Arrastia, R. D., Diringer, M., Figaji, A., Gao, G., Geocadin, R., Ghajar, J., Harris, O., Hoffer, A., Hutchinson, P., Joseph, M., Fouad, K., Bixby, J. L., Callahan, A., Grethe, J. S., Jakeman, L. B., Kitagawa, R., Manley, G., Mayer, S., Menon, D. K., Meyfroidt, Lemmon, V. P., Magnuson, D. S. K., Martone, M. E., Nielson, J. G., Michael, D. B., Oddo, M., Okonkwo, D., Patel, M., L., Schwab, J. M., Taylor-Burds, C., Tetzlaff, W., Torres-Espin, A., Robertson, C., Rosenfeld, J. V., Rubiano, A. M., Sahuquillo, J., Ferguson, A. R., the FAIR-SCI Ahead Workshop Participants, Servadei, F., Shutter, L., Stein, D., Stocchetti, N., Taccone, F. S., Alam,S., Bacon, M., Bambrick,L.,Basso,M.,etal.(2020). Timmons, S., Tsai, E., Ullman, J. S., Vespa, P., Videtta, W., Wright, FAIR SCI ahead: The evolution of the open data commons for D. W., Zammit, C., & Hawryluk, G. W. J. (2020). A management pre-clinical spinal cord injury research. Journal of Neurotrauma, algorithm for adult patients with both brain oxygen and intracranial 37(6), 831–838. https://doi.org/10.1089/neu.2019.6674. pressure monitoring: The Seattle international severe traumatic brain Gøtzsche, P. C., Hróbjartsson, A., Marić, K., & Tendal, B. (2007). Data injury consensus conference (SIBICC). Intensive Care Medicine, extraction errors in meta-analyses that use standardized mean differ- 46(5), 919–929. https://doi.org/10.1007/s00134-019-05900-x. ences. JAMA, 298(4), 430–437. https://doi.org/10.1001/jama.298.4.430. CMAJ. (2014). The “file drawer” phenomenon: Suppressing clinical ev- Hadley, M., Wlaters, B., & Grabb, P. (2002). Guidelines for Management idence. Canadian Medical Association Journal, 170(4), 437. of Acute Cervical Spinal Injuries. Neurosurgery, 50(3), S1. https:// Constantini, S., & Young, W. (1994). The effects of methylprednisolone doi.org/10.1097/00006123-200203001-00003. and the ganglioside GM1 on acute spinal cord injury in rats. Journal Hawkins, B. E., Huie, J. R., Almeida, C., Chen, J., & Ferguson, A. R. of Neurosurgery: Spine, 80(1), 97–111. (2019). Data dissemination: Shortening the long tail of traumatic Crowe, M. J., Bresnahan, J. C., Shuman, S. L., Masters, J. N., & Beattie, brain injury dark data. Journal of Neurotrauma, neu.2018.6192. M. S. (1997). Apoptosis and delayed degeneration after spinal cord https://doi.org/10.1089/neu.2018.6192. injury in rats and monkeys. Nature Medicine, 3(1), 73–76. https:// Hawryluk, G. W. J., Nielson, J. L., Huie, J. R., Zimmermann, L., Saigal, R., doi.org/10.1038/nm0197-73. Ding,Q.,Hirschi,R.,Zeiler,F.A.,Ferguson,A.R.,&Manley,G.T. Dakson, A., Brandman, D., Thibault-Halman, G., & Christie, S. D. (2020). Analysis of Normal High-Frequency Intracranial Pressure (2017). Optimization of the mean arterial pressure and timing of Values and Treatment Threshold in Neurocritical Care Patients: surgical decompression in traumatic spinal cord injury: A retrospec- Insights Into Normal Values and a Potential Treatment Threshold [pub- tive study. Spinal Cord, 55(11), 1033–1038. https://doi.org/10. lished online ahead of print, 2020 Jun 15]. JAMA Neurology, e201310. 1038/sc.2017.52. https://doi.org/10.1001/jamaneurol.2020.1310. Ehsanian, R., Haefeli, J., Quach, N., Bresnahan, J. C., Beattie, M. S., Hawryluk, G., Whetstone, W., Saigal, R., Ferguson, A., Talbott, J., Ferguson, A. R., & McKenna, S. L. (2020). Exploration of surgical Bresnahan, J., Dhall, S., Pan, J., Beattie, M., & Manley, G. (2015). blood pressure management and expected motor recovery in indi- Mean arterial blood pressure correlates with neurological recovery viduals with traumatic spinal cord injury. Spinal Cord, 58,377–386. after human spinal cord injury: Analysis of high frequency physio- Ercan, I. (2015). Examining of published articles with respect to statistical logic data. Journal of Neurotrauma, 32(24), 1958–1967. https://doi. errors in medical sciences. International Journal of Hematology and org/10.1089/neu.2014.3778. Oncology, 25(2), 130–138. https://doi.org/10.4999/uhod.15942. Hook, M. A., Ferguson, A. R., Garcia, G., Washburn, S. N., Koehly, L. Ercan, I., Kaya, M. O., Uzabaci, E., Mankir, S., Can, F. E., & Bashir M., & Grau,J.W.(2004).Monitoring recovery after injury: Albishir, M. (2017). Examination of published articles with respect Procedures for deriving the optimal test window. Journal of to statistical errors in veterinary sciences. Acta Veterinaria, 67(1), Neurotrauma, 21(1), 109–118. https://doi.org/10.1089/ 33–42. https://doi.org/10.1515/acve-2017-0004. 089771504772695995. Faden, A. I. (1989). TRH analog YM-14673 improves outcome follow- Huerta, M. F., Koslow, S. H., & Leshner, A. I. (1993). The human brain ing traumatic brain and spinal cord injury in rats: Dose-response project: An international resource. Trends in Neurosciences, 16(11), studies. Brain Research, 486(2), 228–235. 436–438. https://doi.org/10.1016/0166-2236(93)90069-X. Ferguson, A. R., Christensen, R. N., Gensel, J. C., Miller, B. A., Sun, F., Huie, J. R., Almeida, C. A., & Ferguson, A. R. (2018). Neurotrauma as a Beattie, E. C., Bresnahan, J. C., & Beattie, M. S. (2008). Cell death big-data problem. Current Opinion in Neurology, 31(6), 702–708. after spinal cord injury is exacerbated by rapid TNF -induced traf- https://doi.org/10.1097/WCO.0000000000000614. ficking of GluR2-lacking AMPARs to the plasma membrane. Kaye, J., Heeney, C., Hawkins, N., de Vries, J., & Boddington, P. (2009). Journal of Neuroscience, 28(44), 11391–11400. https://doi.org/10. Data sharing in genomics—Re-shaping scientific practice. Nature 1523/JNEUROSCI.3708-08.2008. Reviews Genetics, 10(5), 331–335. https://doi.org/10.1038/nrg2573. Ferguson, A. R., Hook, M. A., Garcia, G., Bresnahan, J. C., Beattie, M. S., & Kepler, C. K., Schroeder, G. D., Martin, N. D., Vaccaro, A. R., Cohen, M., & Grau, J. W. (2004). A simple post hoc transformation that improves the Weinstein, M. S. (2015). The effect of preexisting hypertension on early metric properties of the BBB scale for rats with moderate to severe neurologic results of patients with an acute spinal cord injury. Spinal spinal cord injury. JournalofNeurotrauma, 21(11), 1601–1613. Cord, 53(10), 763–766. https://doi.org/10.1038/sc.2015.76. Ferguson,A. R.,Irvine,K.-A.,Gensel,J.C.,Nielson,J.L.,Lin, A.,Ly,J., Kigerl, K. A., Gensel, J. C., Ankeny, D. P., Alexander, J. K., Donnelly, D. Segal, M. R., Ratan, R. R., Bresnahan, J. C., & Beattie, M. S. (2013). J., & Popovich, P. G. (2009). Identification of two distinct macro- Derivation of multivariate Syndromic outcome metrics for consistent phage subsets with divergent effects causing either neurotoxicity or testing across multiple models of cervical spinal cord injury in rats. regeneration in the injured mouse spinal cord. Journal of PLoS One, 8(3), e59712. https://doi.org/10.1371/journal.pone.0059712. Neuroscience, 29(43), 13435–13444. https://doi.org/10.1523/ Ferguson, A. R., Nielson, J. L., Cragin, M. H., Bandrowski, A. E., & JNEUROSCI.3257-09.2009. Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B. (2017). Martone, M. E. (2014). Big data from small data: Data-sharing in the “long tail” of neuroscience. Nature Neuroscience, 17(11), 1442– lmerTest package: Tests in linear mixed effects models. Journal of 1447. https://doi.org/10.1038/nn.3838. Statistical Software, 82(13). https://doi.org/10.18637/jss.v082.i13. Lander, E. S. (1996). The new genomics: Global views of biology. Science, Ferguson, A. R., Stück, E. D., & Nielson, J. L. (2011). Syndromics: A bioinformatics approach for Neurotrauma research. Translational 274(5287), 536–539. https://doi.org/10.1126/science.274.5287.536. Lemmon, V. P., Ferguson, A. R., Popovich, P. G., Xu, X.-M., Snow, D. Stroke Research, 2(4), 438–454. https://doi.org/10.1007/s12975-011- 0121-1. M., Igarashi, M., Beattie, C. E., Bixby, J. L., & the MIASCI Consortium. (2014). Minimum information about a spinal cord in- Fisher,M.,Feuerstein,G.,Howells,D.W.,Hurn,P.D.,Kent,T.A.,Savitz,S. I., & Lo, E. H. (2009). Update of the stroke therapy academic industry jury experiment: A proposed reporting standard for spinal cord 52 Neuroinform (2022) 20:39–52 injury experiments. JournalofNeurotrauma,31(15), 1354–1361. What’s there, what’s missing, and what role is there for a global observatory? Lancet, 382(9900), 1286–1307. https://doi.org/10. https://doi.org/10.1089/neu.2014.3400. Lüdecke, D. (2020). sjstats: Collection of Convenient Functions for 1016/S0140-6736(13)61046-6. Common Statistical Computations (0.19.9) [computer software]. RStudio Team. (2018). RsStudio: Integrated development for R. R https://cran.r-project.org/web/packages/sjstats/index.html Studios Inc. http://www.rstudio.com Marmarou, A., Lu, J., Butcher, I., McHugh, G. S., Mushkudiani, N. A., Sabit, B., Zeiler, F. A., & Berrington, N. (2018). The impact of mean Murray,G.D.,Steyerberg,E.W.,&Maas,A.I.R.(2007).IMPACT arterial pressure on functional outcome post trauma-related acute database of traumatic brain injury: Design and description. Journal of spinal cord injury: A scoping systematic review of the human liter- Neurotrauma, 24(2), 239–250. https://doi.org/10.1089/neu.2006.0036. ature. J Intensive Care Med, 33(1), 3–15. https://doi.org/10.1177/ Mueck, L. (2013). Report the awful truth! Nature Nanotechnology, 8(10), 693–695. https://doi.org/10.1038/nnano.2013.204. Shah, A., Almenawer, S., & Hawryluk, G. (2019). Timing of Neff, E. P. (2018). Dark data see the light. Lab Animal, 47(2), 45–48. Decompressive Craniectomy for ischemic stroke and traumatic https://doi.org/10.1038/laban.1405. brain injury: A review. Front Neurol, 10, 11. https://doi.org/10. Nielson, J. L., Cooper, S. R., Seabury, S. A., Luciani, D., Fabio, A., 3389/fneur.2019.00011. Temkin, N. R., & Ferguson, A. R. (2020). Statistical guidelines for Simundic, A.-M., & Nikolac, N. (2009). Statistical errors in manuscripts handling missing data in traumatic brain injury clinical research submitted to Biochemia Medica journal. Biochemia Medica, 19(3), [published online ahead of print, 2020 mar 10]. Journal of 430–437. Neurotrauma. https://doi.org/10.1089/neu.2019.6702. Squair, J. W., Bélanger, L. M., Tsang, A., Ritchie, L., Mac-Thiong, J.-M., Nielson, J. L., Guandique, C. F., Liu, A. W., Burke, D. A., Lash, A. T., Parent, S., Christie, S., Bailey, C., Dhall, S., Charest-Morin, R., Moseanko, R., Hawbecker, S., Strand, S. C., Zdunowski, S., Irvine, Street, J., Ailon, T., Paquette, S., Dea, N., Fisher, C. G., Dvorak, K.-A., Brock, J. H., Nout-Lomas, Y. S., Gensel, J. C., Anderson, K. M. F., West, C. R., & Kwon, B. K. (2019). Empirical targets for D., Segal, M. R., Rosenzweig, E. S., Magnuson, D. S. K., acute hemodynamic management of individuals with spinal cord Whittemore,S.R., McTigue, D.M.,Popovich,P. G., injury. Neurology, 93(12), e1205–e1211. https://doi.org/10.1212/ Rabchevsky, A. G., Scheff, S. W., Steward, O., Courtine, G., WNL.0000000000008125. Edgerton, V. R., Tuszynski, M. H., Beattie, M. S., Bresnahan, J. Walters,B.C., Hadley, M.N., Hurlbert,R.J.,Aarabi, B., Dhall, S. S., Gelb, C., & Ferguson, A. R. (2014). Development of a database for trans- D. E., Harrigan, M. R., Rozelle, C. J., Ryken, T. C., & Theodore, N. lational spinal cord injury research. Journal of Neurotrauma, (2013). Guidelines for the Management of Acute Cervical Spine and 31(21), 1789–1799. https://doi.org/10.1089/neu.2014.3399. Spinal Cord Injuries: 2013 update. Clin Neurosurg, 60,10. Nielson, J. L., Paquette, J., Liu, A. W., Guandique, C. F., Tovar, C. A., Wilkinson, M. D., Dumontier, M., Aalbersberg, I. J., Appleton, G., Inoue, T., Irvine, K.-A., Gensel, J. C., Kloke, J., Petrossian, T. C., Axton, M., Baak, A., Blomberg, N., Boiten, J.-W., da Silva Lum, P. Y., Carlsson, G. E., Manley, G. T., Young, W., Beattie, M. Santos, L. B., Bourne, P. E., Bouwman, J., Brookes, A. J., Clark, S., Bresnahan, J. C., & Ferguson, A. R. (2015). Topological data T., Crosas, M., Dillo, I., Dumon, O., Edmunds, S., Evelo, C. T., analysis for discovery in preclinical spinal cord injury and traumatic Finkers,R.,Gonzalez-Beltran,A.,Gray,A. J.G.,Groth,P., brain injury. Nature Communications, 6(1), 8581. https://doi.org/10. Goble, C., Grethe, J. S., Heringa, J., ’t Hoen, P. A. C., Hooft, R., 1038/ncomms9581. Kuhn, T., Kok, R., Kok, J., Lusher, S. J., Martone, M. E., Mons, A., NIH Office of Extramural Research. (2019). NIH Grants Policy Packer, A. L., Persson, B., Rocca-Serra, P., Roos, M., van Schaik, Statement, 8.4.2 Record Retention and Access [Government]. R., Sansone, S. A., Schultes, E., Sengstag, T., Slater, T., Strawn, G., Nih.Gov/. https://grants.nih.gov/grants/policy/nihgps/html5/ Swertz, M. A., Thompson, M., van der Lei, J., van Mulligen, E., section_8/8.4.2_record_retention_and_access.htm Velterop, J., Waagmeester, A., Wittenburg, P., Wolstencroft, K., Nout,Y.S.,Beattie,M.S.,&Bresnahan,J.C.(2012).Severityof Locomotor Zhao, J., & Mons, B. (2016). The FAIR guiding principles for sci- and cardiovascular derangements after experimental high-thoracic spi- entific data management and stewardship. Scientific Data, 3(1), nal cord injury is anesthesia dependent in rats. J Neurotrauma, 29(5), 160018. https://doi.org/10.1038/sdata.2016.18. 990–999. https://doi.org/10.1089/neu.2011.1845. Young, W. (2002). Chapter 17 spinal cord contusion models. In Progress Parikshak, N. N., Gandal, M. J., & Geschwind, D. H. (2015). Systems in Brain Research (Vol. 137, pp. 231–255). Elsevier. https://doi.org/ biology and gene networks in neurodevelopmental and neurodegen- 10.1016/S0079-6123(02)37019-5. erative disorders. Nat Rev Genet, 16(8), 441–458. https://doi.org/10. Yue,J.K.,Hemmerle,D.D.,Winkler,E.A.,Thomas,L.H., Fernandez, X. 1038/nrg3934. D.,Kyritsis,N.,Pan,J. Z.,Pascual,L.U.,Singh,V.,Weinstein,P.R., R Core Team. (2019). R: A language and environment for statistical comput- Talbott,J.F.,Huie,J.R.,Ferguson,A.R.,Whetstone,W.D.,Manley, ing. R Foundation for Statistical Computing. https://www.R-project.org/ G. T., Beattie, M. S., Bresnahan, J. C., Mummaneni, P. V., & Dhall, S. Readdy, W. J., Saigal, R., Whetstone, W. D., Mefford, A. N., Ferguson, S. (2020). Clinical Implementation of Novel Spinal Cord Perfusion A. R., Talbott, J. F., Inoue, T., Bresnahan, J. C., Beattie, M. S., Pan, Pressure Protocol in Acute Traumatic Spinal Cord Injury at U.S. J., Manley, G. T., & Dhall, S. S. (2016). Failure of mean arterial Level I Trauma Center: TRACK-SCI Study. World Neurosurgery, pressure goals to improve outcomes following penetrating spinal 133, e391–e396. https://doi.org/10.1016/j.wneu.2019.09.044. cord injury. Neurosurgery, 79(5), 708–714. https://doi.org/10. Yue,J.K.,Winkler,E.A.,Rick,J.W.,Deng, H.,Partow,C.P., 1227/NEU.0000000000001249. Upadhyayula, P. S., Birk, H. S., Chan, A. K., & Dhall, S. S. Rocher, L., Hendrickx, J. M., & de Montjoye, Y.-A. (2019). Estimating (2017). Update on critical care for acute spinal cord injury in the the success of re-identifications in incomplete datasets using gener- setting of polytrauma. Neurosurgical Focus, 43(5), E19. https://doi. ative models. Nat Commun, 10(1), 3069. https://doi.org/10.1038/ org/10.3171/2017.7.FOCUS17396. s41467-019-10933-3. Røttingen, J.-A., Regmi, S., Eide, M., Young, A. J., Viergever, R. F., Årdal, C., Guzman, J., Edwards, D., Matlin, S. A., & Terry, R. F. Publisher’sNote Springer Nature remains neutral with regard to jurisdic- (2013). Mapping of available health research and development data: tional claims in published maps and institutional affiliations.
Neuroinformatics – Springer Journals
Published: Jan 1, 2022
Keywords: Data science; Metascience; Neurotrauma; Reproducibility; Spinal contusion; Motor recovery; Autonomic; Hemodynamics
Access the full text.
Sign up today, get DeepDyve free for 14 days.