Open Access Platforms in Spinal Cord Injury: Existing Clinical Trial Data to Predict and Improve Outcomes

Abstract

Recovery from acute spinal cord injury (SCI) is characterized by extensive heterogeneity, resulting in uncertain prognosis. Reliable prediction of recovery in the acute phase benefits patients and their families directly, as well as improves the likelihood of detecting efficacy in clinical trials. This issue of heterogeneity is not unique to SCI. In fields such as traumatic brain injury, Parkinson’s disease, and amyotrophic lateral sclerosis, one approach to understand variability in recovery has been to make clinical trial data widely available to the greater research community. We contend that the SCI community should adopt a similar approach in providing open access clinical trial data.

Recovery from acute spinal cord injury (SCI) is characterized by extensive heterogeneity. This is true for both standard neurological outcomes (eg, muscle strength and sensation) and functional abilities (eg, ambulation). Our ability to predict the extent of recovery from observations early post injury is, unfortunately, extremely limited. Only a handful of subject demographics and injury characteristics have consistently demonstrated strong prognostic value (eg, sensory sparing). ¹ As a result, patients recover to vastly different levels, even though they may appear similar initially. This poses a clinical problem, because patients with SCI have urgent questions early after injury: Will I regain any function in my hands? Can I return to work? At present, these questions are difficult to answer with any certainty before 6 months post-SCI. In addition, variable recovery poses challenges to the design and analysis of clinical trials: How is therapeutic efficacy measured on a background of such extreme heterogeneity? Many consider recovery variability a major reason why clinical trials have thus far failed to detect significant treatment effects. ² Reliable prediction of recovery in the acute phase benefits patients directly and improves the likelihood of detecting efficacy in clinical trials.

Heterogeneity among individual trajectories of recovery is not unique to SCI. Other traumatic insults to the nervous system (eg, traumatic brain injury [TBI]) and neurodegenerative conditions, such as Parkinson’s disease (PD) and amyotrophic lateral sclerosis (ALS), are also subject to considerable variability in disease progression. What are these fields doing to understand the heterogeneity of their conditions? One approach has been to make clinical trial data widely available to the greater research community, for the essential purpose of characterizing disease progression.

The TBI community has made enormous progress in sharing clinical trial data, beginning with the development of the International Mission for Prognosis and Analysis of Clinical Trials (IMPACT; tbi-impact.org), which was designed to capture all clinical trial data collected at the time of its advent. This process has evolved into large international initiatives, such as the International Traumatic Brain Injury Research (InTBIR) cooperative effort (intbir.nih.gov), which includes a platform to facilitate data sharing across studies (the Federal Interagency TBI Research Informatics System; fitbir.nih.gov). Funding agencies, such as the National Institute of Neurological Disorders and Stroke, have even begun to mandate data sharing as a requirement to receive support (ninds.nih.gov/research/clinical_research/toolkit/archived_datasets.htm).

For clinical trials conducted in ALS, the Neurological Clinical Research Institute at Massachusetts General Hospital created the Pooled Resource Open-Access ALS Clinical Trials platform (PRO-ACT; ALSdatabase.org/). ³ The vision of the PRO-ACT is to “accelerate and enhance translational ALS research” by designing and building a platform that contains the merged data from as many completed ALS clinical trials as possible. Launched in December 2012, the PRO-ACT database now contains data on more than 10 000 patients from 23 Phase II/III clinical trials.

After developing PRO-ACT, these pioneers turned to crowdsourcing to mine the aggregated clinical trial data. ³ To address the issue of variability in ALS, a subset of the PRO-ACT data was used before its public launch for an international crowdsourcing initiative: The DREAM-Phil Bowen ALS Prediction Prize4Life. The challenge awarded US$50 000 for the most accurate methods to predict ALS progression. In this challenge, solvers were asked to use 3 months of baseline clinical trial data to predict disease progression over the subsequent 9 months. The challenge garnered submission of 37 unique algorithms from which 2 winning entries were identified. The results of this novel crowdsourcing initiative were recently published in the prestigious journal Nature Biotechnology. ³ Since the challenge, several other studies using PRO-ACT data have revealed previously unknown predictors of disease progression.^4–6 A similar crowdsourcing challenge has been issued for Parkinson’s disease (PPMI-info.org).

By comparison, clinical SCI research is lagging behind. This shortcoming cannot be attributed to a lack of completed clinical trials. Indeed, a number of very prominent studies are well suited for inclusion in a dataset to perform secondary analyses. The National Spinal Cord Injury Studies (NASCIS I, II, and III), randomized nearly 1000 patients, ranging from mild (ie, sensory impairments only) to severe traumatic SCI (ie, complete sensory loss and motor paralysis). ⁷ The Sygen multicenter study enrolled 760 subjects with acute traumatic SCI.^8,9 Trials such as NASCIS and Sygen remain valuable resources to examine progression because of their size and the incorporation of common neurological data elements. ¹⁰ To some degree, researchers have accessed these clinical trial data (including the authors) ¹ ; however, this access is limited. A number of other clinical trials have been completed in the field of SCI since these seminal studies, and others are currently underway. ¹¹ However, clinical trial data sets in SCI, by and large, remain inaccessible to the wider scientific community.

The clinical SCI world is even falling behind our preclinical colleagues. A group at the University of California–San Francisco has initiated an international platform to share data from experimental SCI studies in rodents. ¹² Based on observations from thousands of animals from 60 preclinical studies, and through the application of unbiased statistical approaches, these investigators have elucidated observations regarding recovery after experimental injuries, and illustrated novel effects due to intervention. ¹² Sharing data that would otherwise be discarded (or lost) has been key to the success of this project. The same can and should be done for clinical trial data in humans.

The field of SCI has made some progress in understanding the natural progression of injury. Existing initiatives designed for this very purpose include the widely successful European Multi-Center Study about Spinal Cord Injury (EMSCI; emsci.org). Impressively, the EMSCI continues to be run with almost no support, highlighting the success of this massive collaboration. There are, however, a number of notable limitations to such an observational approach. First, assessments at very acute time-points are difficult to reliably capture. For example, the earliest available time-point in the EMSCI is at 1 to 2 weeks postinjury. In contrast, since many therapies need to be initiated very early after injury, clinical trials often incorporate collection of outcome measures in very acute stages. In addition, subject dropout hampers longitudinal observational cohort studies. This raises a concern that dropout is potentially biasing our knowledge of “natural” recovery, currently based on observational studies. By comparison, in a clinical trial, there is a greater incentive (and funding) to acquire measures at all time-points. Finally, clinical trials typically capture information beyond the scope of a conventional observational study, including data on concomitant medications, adverse events, blood chemistry, and surgical details.

Initiating an open access clinical trial data repository needs to overcome some obstacles unique to humans. ¹³ Luckily, these have been largely addressed before us. Secondary use of clinical trial data is already permissible without informed consent from the original participants if the data are completely anonymized.^14-17 Anonymizing the data can be achieved by removing identifying details. ¹⁸ In PRO-ACT, anonymization was achieved by removing patient names or initials, dates of birth (age at enrollment was provided instead), and trial-specific information (eg, trial location, center identity, trial dates, and drug being tested), as well as creating new subject IDs. ¹⁹

Intellectual property issues will undoubtedly be another major concern for academic and industry partners. These too can be mitigated. First, as a secondary analyst (or “data parasite” ²⁰ ), providing “due credit” may warrant sharing authorship with the original investigators (see National Institutes of Health “Guiding principles for sharing clinical trial data”^14-16). In addition to providing incentives, this may have a number of indirect benefits, including assurances of data validity. Second, through the removal of study elements in the shared data (eg, drug assignment group, study site), it would be impossible to re-analyze for treatment effects or to examine differences between study centers. Above all else, intellectual property issues may be overcome by the incentive of a data-rich resource to inform future clinical trials.

A final challenge is deciding who will ultimately be responsible for maintaining and dispensing the data. Encompassed in this challenge are financial costs (eg, data storage) and personnel to meet the demands of data requests. Many key partners in the field of SCI will need to be involved in this dialogue, including academics, clinicians, funding agencies, and industry, as well as colleagues from outside of SCI who have already managed all of the hurdles to data sharing (eg, TBI community).

Regardless of the costs, the risks of NOT centralizing SCI clinical trial data for open access are extreme. Both the NASCIS and Sygen trials are over 20 years old. With time, old data will degrade, and eventually be lost or cease to exist in a useable format (eg, with undefined variables or outdated file types). Losing data that could serve a valuable role in the prediction of recovery (and in turn improve future trials) only compounds the disappointment that these studies ultimately failed to improve outcomes. An open access platform encourages participation by a wider scientific community and acknowledges that our own scientific biases may limit our view of SCI. By adopting an open access approach and analyzing clinical trial data as a collective unit, we may uncover a litany of unknowns that expand our understanding of recovery after SCI. Establishing a platform for historical clinical trials will also develop the infrastructure for trials emerging in the future. It is time for the SCI community to be proactive.