MICHAEL B. BRACKEN, PH.D., AND THEODORE R. HOLFORD, PH.D.

Transcription

1 See the Editorial and the Response in this issue, pp J Neurosurg (Spine 3) 96: , 2002 Neurological and functional status 1 year after acute spinal cord injury: estimates of functional recovery in National Acute Spinal Cord Injury Study II from results modeled in National Acute Spinal Cord Injury Study III MICHAEL B. BRACKEN, PH.D., AND THEODORE R. HOLFORD, PH.D. Departments of Epidemiology and Public Health, Neurology, and Statistics, School of Medicine, and Graduate School, Yale University, New Haven, Connecticut Object. In the second National Acute Spinal Cord Injury Study (NASCIS II) investigators evaluated several standard neurological parameters but not functional activity. This has led to questions concerning the clinical importance of the increase in neurological recovery observed following administration of methylprednisolone (MP) within 8 hours of acute spinal cord injury (SCI). The safety of the therapy has also been questioned. Methods. Both neurological and functional recovery were assessed in NASCIS III, a trial that followed an almost identical protocol to NASCIS II. In the current analysis locally weighted scatterplot smoothing (LOESS) nonparametric regression is used to model the extent of recovery in the Functional Independence Measure (FIM) that is predicted by improvement in the NASCIS/American Spinal Cord Injury Association motor scores that were documented in NASCIS III 1 year after SCI, and the models are applied to the extent of motor recovery demonstrated in NASCIS II. The models predict improvement in FIM that would be expected from the motor function recovery observed in NASCIS II. Estimates are provided overall and for patients with complete and incomplete neurological loss at time of injury. The authors review recent evidence obtained from randomized studies documenting adverse effects that may result from high-dose MP therapy. The relationship between motor function and FIM is strongly nonlinear and dependent on initial level of injury and degree of injury severity. In the best statistical model, the expanded motor score could be used to explain 77.2% of the variability in the FIM. Based on the mean MP-related 3.6-unit improvement in the motor score for patients with complete injuries and 7.3 for those with incomplete injuries owed to MP in NASCIS II, 18.6% of patients would improve six or more FIM points and 9% nine or more points, respectively. In those with complete neurological injury, the mean motor improvement of 3.6 predicted that 63.9% of the patients would improve three or more FIM points and 12.1% six or more points to a maximum of eight points. Of those with incomplete neurological injury, a 7.3 mean improvement in motor function predicted that 27.4% would gain six or more FIM points and that 21% would gain nine or more points to a maximum of 15 points. Analysis of the current best evidence from SCI and other randomized surgical trials in which highdose MP has been administered provides no grounds for concern about commonly studied adverse effects. Conclusions. The extent of MP therapy related motor function recovery observed in NASCIS II predicted clinically important recovery in the FIM. Reasons to be cautious with regard to this prediction include the lack of robustness in statistical modeling, some loss of validity in the FIM, and considerable heterogeneity in the SCI population. Whatever functional activity is ascribed to high-dose MP therapy, it is does not appear to be associated with risk of adverse outcomes. KEY WORDS methylprednisolone motor function neurological function National Acute Spinal Cord Injury Study D Abbreviations used in this paper: ASIA = American Spinal Injury Association; FIM = Functional Independence Measure; IMSOP = International Medical Society of Paraplegia; LOESS = locally weighted scatterplot smoothing; MP = methylprednisolone; NASCIS = National Acute Spinal Cord Injury Study; SCI = spinal cord injury. J. Neurosurg: Spine / Volume 96 / April, 2002 URING the acute phase of SCI, the clinical status related to the cord is evaluated using measures of neurological function: typically, motor function and pinprick and light touch sensation. During rehabilitation, however, it is more common to evaluate activities of daily living including measures of self care, transfer (for example, to a bed or wheelchair), and ambulation. Over the last decade there has been considerable agreement with regard to how neurological and functional status should be assessed, with increasing uniformity in methodologies as adopted by international organizations. The relationship between the two most common methods of assessment, however, the ASIA/IMSOP procedures for evaluating neurological function 1 and the FIM for functional status, 25 remains poorly understood. The NASCIS group has used measures that are exactly like (sensory function) or very similar (motor function) to the ASIA/IMSOP methodologies in all its trials; however, the FIM has only been used in NASCIS III. In this paper 259

2 M. B. Bracken and T. R. Holford we analyze data derived from the latter trial 7,8 to investigate the relation of neurological function to the FIM. Using a statistical model derived from the NASCIS III data, we apply the model to neurological scores obtained in NASCIS II 5,6 to obtain an estimate of the degree of functional recovery that would be predicted from use of MP in that trial. Clinical Material and Methods Patient Population The patient population is comprised of patients treated in NASCIS III 7,8 and NASCIS II, 5,6 both randomized controlled trials evaluating the efficacy of pharmacological therapies in the early hours postinjury. Standardized protocols for methods, timing, and qualifications of examiners, as well as for data collection and quality assurance, were in place throughout the trials. Our data are taken from the baseline (hospital admission) and 1-year ( days from injury) follow-up examinations in both trials. Examinations were only performed by study personnel, all of whom had observed specifically designed training tapes or undergone hands-on training. Examiners were also credentialed in the FIM examination. 25 Definition of Study Variables Following are brief operational definitions of variables used in the present analysis. Motor Assessment. Fourteen muscle roots were measured bilaterally as follows: 0, no contraction; 1, flicker/ trace of contraction; 2, active movement without antigravity; 3, active movement with antigravity; 4, active movement against resistance; and 5, normal. The expanded score ranged from 0, which indicated no motor activity, to 70, which indicated normal motor function. Sensory Assessment. Twenty-nine spinal cord segments were independently evaluated using pinprick and light touch and were scored as follows: 1, sensation absent; 2, dysfunctional (including hypesthesia); and 3, normal. Each expanded score ranged from 29, which indicated no response in any segment, to 87, which indicated that all segments were normal. Deep Pain Assessment. Six parameters (wrist, thumb, little finger, knee, ankle, and great toe) were tested for deep pain/pressure sensation and recorded bilaterally as: 1, absent; 2, decreased; or 3, normal. The expanded score ranged from 6, which indicated completely absent sensation, to 18, which denoted completely normal status. Functional Independence Measure. The FIM measures 18 items in six domains: self-care, sphincter control, mobility, locomotion, communication, and social cognition; it provides an overall score of function ranging from 18, indicating the need for assistance in all areas, to 126, indicating complete independence. Patients are grouped into one of seven categories based on the extent of their need; this ranges from whether they require total assistance/ maximal assistance (help is needed but patient may assist 25% of the time), moderate or minimal assistance (patient can function between 50 74%, or 75% or more of the time, or needs supervision), modified independence (no need for assistance but requires devices), and complete independence. The communication and social cognition FIM domains were not used in this analysis because they were observed to be generally unaffected by SCI in our original analysis 7 and in the analyses reported in the present paper. The total score of the FIM without these items ranged from 13 to 91. The estimated effect of MP on the motor score was obtained from the NASCIS II; we used the expanded motor function score. We selected patients in whom the time from SCI to the onset of treatment was within 8 hours of the injury (a value chosen because it was close to the median time to treatment demonstrated in that study and the time period shown in NASCIS II when MP is efficacious). 5,6 Statistical Analysis Analysis of covariance provided estimates of the effect of treatment on the change in score from baseline, yielding estimates for those receiving treatment within 8 hours. In addition, this analysis was stratified by whether the injury was diagnosed as complete or incomplete in the emergency room. Both stratifications were planned in the study protocol. To create a model for the association between motor score and FIM, we used results from NASCIS III, in which both expanded motor scores and the FIMs were determined in each patient. A model was then used to estimate the FIMs for individuals in the placebo-treated group in NASCIS II, as well as the corresponding FIMs that would have resulted if the patients expanded scores had been equal to the mean estimated effect from this study. The difference provided an estimate of the effect of MP treatment on the FIM in NASCIS II. A more detailed account of these statistical methods has been provided (see Statistical Appendix). Results A summary of the neurological and functional variables used in the analysis is provided in Table 1. Of 499 patients randomized to NASCIS III, 431 were alive and underwent 1-year follow-up examination; 8 of these 430 successfully completed an FIM and 429 a neurological examination. Discomfort or other medical emergency precluded completion of FIM or neurological exam in three patients. The level of dysfunction experienced at 1 year is also detailed in Table 1. The comprehension and expression domains of the FIM were shown to be normal. In Table 2 we report the Spearman rank correlations between 1-year neurological scores and 1-year FIM scores. Overall, motor scores showed higher correlation with each FIM score than did any other neurological parameter. Pinprick sensation score was the second most highly correlated variable followed by touch and deep pain sensation. The motor score correlations ranged from (with eating) to (climbing stairs). Not unexpectedly, measures of comprehension and expression showed very low correlations with all neurological parameters and were eliminated from the analyzed FIM subset. In the first step in the analysis we used information obtained from NASCIS III to develop a regression calibration model with which to predict FIM scores from the expanded motor scores. Figure 1 shows a scatter plot of 260 J. Neurosurg: Spine / Volume 96 / April, 2002

3 Neurological and functional status after SCI TABLE 1 Summary of neurological and functional variables used in analysis of NASCIS III data No. of Mini- Maxi- Variable Patients Mean SD* mum mum expanded score at 1 yr motor pinprick touch deep pain/pressure self care eating grooming bathing dressing upper body dressing lower body toileting sphincter control bladder bowel transfer bed, chair, wheelchair toilet tub, shower walking or use of a wheelchair climbing stairs comprehension expression * SD = standard deviation. the relationship between the subset of FIM scores. Overall, there was a strong nonlinear relationship between the expanded motor score at 1 year postinjury and the corresponding FIM score. The lack of linearity resulted in large part from a lack of association with the FIM score in patients with 1-year motor scores between 35 and 60. A LOESS curve predicting FIM from the expanded motor score at 1 year accounted for (R 2 ) of the variability in the FIM (Fig. 1). Also shown in Fig. 1 are 1-year FIM estimates for patients with complete and incomplete neurological injuries determined at baseline, as well as the distribution of baseline motor scores. The bar graphs in Fig. 1 show the NASCIS III distribution of 1-year motor scores by whether the injury was complete or incomplete. Cases with complete injuries provide most of the observations covering the full range of scores at 1-year follow up. Evidence of a linear relationship was more apparent in patients with incomplete injury but the number of observations was much smaller, especially in cases of motor scores lower than 30. Therefore, we used the single curve, based on all the data, as the regression calibration model for estimating the expected FIM. In additional analyses (not shown), we explored associations between FIM and expanded scores for pinprick, touch, and pain sensation, but none of these was as strong as the motor score in predicting FIM, nor did they improve the prediction based on motor scores alone; thus, they are not investigated further in this communication. The nonlinearity of the relationship shown in Fig. 1 implies that the effect of treatment on FIM only depends in part on the motor score. Figure 2 shows the estimated change in FIM that was expected based on a 5.1 improvement in motor score, the estimated effect of MP treatment J. Neurosurg: Spine / Volume 96 / April, 2002 in all patients in NASCIS II, and the effects of motor improvement of 3.6 (complete neurologically injured cases) and 7.3 units (incomplete neurologically injured cases). Clearly, the greatest treatment effect related impact of a given size will occur in individuals with motor scores in the range of 0 to 30 and 55 to 62. On the other hand, little treatment-related impact on FIM was demonstrated in patients with scores ranging from 40 to 50. Also shown in Fig. 2 is the distribution of motor scores observed at 1 year for placebo-treated patients with complete and incomplete injuries in NASCIS II. Most patients with motor scores in the 0 to 30 range had sustained complete injuries, and smaller treatment effects on the motor scores were observed. Among those with incomplete injuries, the mode was at 70, indicating placebo-treated patients with complete recovery in whom additional benefit from MP treatment was thus not attainable. Table 3 provides a summary of the muscle roots involved in the individual components of the NASCIS/ ASIA score as well as the corresponding spinal cord segments. In earlier work 14 we partitioned the expanded motor score into 1) level of injury and 2) the severity of injury at and below that level. Approximately 80% of patients had sustained a maximum-severity injury; hence, a cumulative sum of the maximum score provides an indication of the injury site. It appears that thoracic injuries resulted in the least estimated improvement in the FIM, whereas in cases of cervical injuries we might expect a greater change in FIM from a specified change in motor score. Overall, 45.9% of patients with an MP-related 5.1 motor point improvement would gain six or more FIM points and 11.6% nine or more FIM points. Figure 3 provides histograms of the estimated mean change in predicted FIM based on the motor scores observed in placebo-treated patients from NASCIS II (stratified by complete or incomplete injury). In patients with complete injuries, the mean estimated FIM improvement produced by a 3.6-unit increase in motor score was 4.4, although the distribution appeared to be bimodal, with one mode of approximately two, and a second of five. In patients with incomplete injuries, a motor improvement of 7.3 would be expected to produce a mode for FIM improvement of 0, but this group consisted entirely of those placebo-treated patients with a perfect score of 70 at 1 year. The range of estimated improvement in FIM scores among remaining patients was 1 to 15, with a mean of 6.5 (3.9, if those placebo-treated patients with normal scores are included). As shown in Fig. 3, of patients with complete injury there were only two in whom no gains in the FIM were documented and there appeared to be two distinct patterns of recovery: 30 (36.1%) of 83 patients improved two or three FIM points and 46 (55.4%) improved five or more FIM points. Gains of more than eight FIM points were not demonstrated in patients with complete injury. Some of the patients (25 [40.3%] of 62) with incomplete injury in the placebo-treated group made a full recovery at 1 year, and we could not estimate a FIM change in them. Among all patients with incomplete injury, a treatment-related gain of six or more FIM points and a gain of nine or more (to a maximum gain of 15 [one patient]) was demonstrated in 13 patients (21%). As shown in Table 4 our analysis-derived estimate was that 13.6% of patients with quadriplegia gained six or 261

4 M. B. Bracken and T. R. Holford TABLE 2 Spearman rank correlation coefficients for 1-year postinjury neurological scores and FIMs from NASCIS III* Neurological Scores at 1 Year (p value) FIM Domain Motor Pinprick Touch Deep Pain eating (0.0001) (0.0001) (0.0001) (0.0001) grooming (0.0001) (0.0001) (0.0001) (0.0001) bathing (0.0001) (0.0001).0583 (0.0001) (0.0001) dressing upper body (0.0001) (0.0001) (0.0001) (0.0001) lower body (0.0001) (0.0001) (0.0001) (0.0001) toileting (0.0001) (0.0001) (0.0001) (0.0001) sphincter bladder (0.0001) (0.0001) (0.0001) (0.0001) bowel (0.0001) (0.0001) (0.0001) (0.0001) transfer bed (0.0001) (0.0001) (0.0001) (0.0001) toilet (0.0001) (0.0001) (0.0001) (0.0001) tub (0.0001) (0.0001) (0.0001) (0.0001) locomotion (0.0001) (0.0001) (0.0001) (0.0001) stairs (0.0001) (0.0001) (0.0001) (0.0001) comprehension (0.0302) (0.3515) (0.5597) (0.6497) expression (0.0019) (0.0314) (0.0552) (0.0615) * The number of patients for each correlation ranged from 427 to 428. more FIM points as a result of MP therapy. The respective percentages for patients with paraplegia, quadriparesis, and paraparesis at 1 year were 20.9, 45.5, and 50%. Discussion Unlike many medical conditions in which the severity of disease processes is measured directly (for example, by histology, tumor staging, or genotyping), acute SCI continues to be measured indirectly. Even when the spinal cord is visualized by radiography or magnetic resonance imaging, this is usually to aid in establishing diagnosis, and current magnetic resonance imaging findings add little to the neurological examination as an adjunct in prognosis. 23 There is no other quantification of anatomical, physiological, or cellular damage that enjoys common acceptance or practice. Spinal cord trauma continues to be assessed indirectly by examination of residual neurological function (typically motor function and pinprick, light touch, and deep pain sensation) with the same measures repeated at postinjury intervals to determine the extent of recovery. Postinjury functional status is also measured to assess the degree of recovery, and there has been considerable effort to standardize these outcomes in recent years, although they continue to suffer some problems of validity and generalizability. 1,17 Measures of neurological function and functional status are related to each other in complex but still poorly understood ways. They have some relationship to each other as evidenced by the fact that both generally either improve or worsen together, statistically one predicts the other, and yet their relationship is far from perfect. In earlier work, we have shown that the relationship depends on how neurological function is categorized, which measure of neurological function is considered, and which activities of daily living are being analysed. Ambulatory and self-care activities, for example, were particularly dependent on the combined and additive effects of motor and sensory functions. 4 In the current analysis because we found that sensory measures added little to the statistical models, we focused on motor function recovery, which dominated the relationship with functional activity, as others have reported. 10 This is not to say that sensation is unimportant in functional recovery, but rather its added effects are difficult to identify in the statistical procedures used here. There have been few attempts to compare neurological status and functional status, 16 and one analysis was seriously flawed when the authors reported a strong negative correlation between the two, almost identical ASIA and NASCIS, measures. 2 In our study, data obtained from the NASCIS III, in which investigators measured both neurological function by using ASIA/IMSOP criteria and functional status by using the FIM, were analyzed to determine further the relationship between these two parameters. The advantage of this dataset over others previously used is that all measurements were conducted using strict protocol devised for purposes of the clinical trials and using specially trained examiners. We also took advantage of new statistical procedures that were unavailable to earlier analysts. Our analysis shows an even more complex relationship between neurological function and functional activity than previously thought. Previous work has suggested that the relationship between these two variables would be linear although of differing strength, 4 but in our new analysis we found that interpretation to be incorrect. To be sure, there are strong linear components, particularly demonstrated in cases of cervical and lower lumbar injuries. Overall, however, there were substantial degrees of nonlinearity, especially in patients with thoracic injuries in whom neurological recovery poorly predicts functional recovery. This complexity has made our second goal in this analysis even more difficult. Publication of the NASCIS II led to questions about the clinical utility of results that were 262 J. Neurosurg: Spine / Volume 96 / April, 2002

5 Neurological and functional status after SCI FIG. 1. Upper: Scatter plot and the LOESS-fitted lines for the relationship between 1-year FIM and motor scores for all patients in NASCIS III by whether the injury was complete at baseline. Circles denote cases of complete injury; triangles, cases of incomplete injury. Center and Lower: Bar graphs showing 1-year FIM estimates for patient with complete (center) and incomplete injury (lower). restricted to neurological function. Although there was little agreement on the preferred method of measuring functional activity when NASCIS II was designed, in hindsight it would have been preferable to use some direct measure of functional status. In this paper we have undertaken an ex post facto analysis by using models of the relationship between neurological and functional statuses derived from the NASCIS III dataset to estimate what degree of improvement in the FIM would be expected from the observed neurological improvement demonstrated in those MP-treated patients in NASCIS II. The FIM itself is an imperfect index to associate with neurological function because it measures disability rather than actual impairment that is, it assesses what people actually do, not what they are capable of. 1 Therefore, it is influenced by factors that may be extraneous to the study of interest such as depression, older age, and other infirmities. The FIM may also be influenced by the relativity of architectural conveniences of individual homes: persons of the same functional ability but having a properly equipped bathroom may be assessed as functioning at higher levels J. Neurosurg: Spine / Volume 96 / April, 2002 than those with poorer-equipped facilities. Because of randomization, these potential confounding factors should not influence the comparisons between MP- and placebo-treated patients. They will, however, reduce the relationship of FIM with the neurological scores. Because the relationship between neurological and functional status was neither a linear nor simple polynomial one, the LOESS estimator was used, but we also fit a number of alternate nonparametric regression models, including cubic splines representing a local third-degree polynomial function (data not shown). The use of some models led to very little difference in the estimates of FIM change from those shown in Fig. 1, but others did result in somewhat different estimates of change in the slope, particularly in the region of a 1-year motor score above 40. This model-related instability cannot be completely explained, but it does reflect random variation about the fitted line, as well as the complex nature of the relationship between the two outcome measures, particularly in less severely injured patients. We also noted that the motor improvement score only measures one (the right) side of the body, a 263

6 M. B. Bracken and T. R. Holford TABLE 3 Cumulative maximum expanded NASCIS motor scores categorized by muscle root Score Maxi- Cumulative Muscle Root Spinal Segment mum Maximum deltoid C biceps C-5, C extensor carpi radialis longus C-6, C triceps C extensor digitorum C-7, C opponens pollicis C-8, T st dorsal interosseus C-8, T iliopsoas L-1, L-2, L quadriceps femoris L-2, L-3, L tibialis anterior L-4, L extensor hallucis longus L-5, S peroneus longus & brevis L-5, S hamstring muscles L-4, L-5, S-1, S gastrocnemius S-1, S FIG. 2. Upper: Estimated change in the FIM produced by the overall treatment effect (5.1 units), the effect in patient with complete injury (3.6 units), and the effect in patients with incomplete injury (7.3 units). Center and Lower: Distribution of 1-year motor scores for placebo-treated patients in NASCIS II stratified by complete (center) or incomplete injuries (lower) determined at baseline examination. strategy followed in all NASCIS analyses to avoid overestimating recovery, by summing two highly correlate outcomes that occurs if both sides are summed. In the current analysis, however, use of one side underestimates the ability of the motor improvement score to predict FIM recovery. Other limitations to this type of analysis also need to be considered. Assumptions must be made when taking data from one trial and applying them to another even when the two trials followed virtually identical protocols and the same population of patients was studied. Moreover, even though these are quite large trials in the context of neurosurgical studies, the number of patients within some subgroups is small and the results reflect a lack of precision. Although we analyzed trial data, the prediction of functional status based on neurological function is not a randomized comparison and small numbers can be problematic. Given these limitations, the present analysis should not be considered as a primary or even secondary outcome of the NACSIS II trial; rather, it is an attempt to estimate functional recovery. The results of NASCIS II continue to be reflected solely in the two primary reports of that trial. 5,6 Our analysis shows that a quite mixed pattern of MPinduced recovery would have been expected in the FIM. Approximately 17.2% of patients would experience no improvement in FIM, and 47.6% would experience improvement of fewer than three points. On the other hand, recovery of six or more points was demonstrated in 18.6% and nine or more points in 9%, which is equivalent to a 10% recovery in the modified total FIM used in this study. In patients with complete injury, a gain of six or more FIM points was demonstrated in 12.1%, and in those with incomplete injury a gain of six or more was observed in 27.4% and a gain of nine or more FIM points in 21%. How should we interpret the clinical importance of this type of change in the FIM? The FIM is scored in gross categories and the actual degree of patient-related benefit would depend on where recovery occurred. In the unlikely event that only one parameter was affected, a six-point recovery in eating, for example (which is the first item in the self-care domain), would move a patient from total assistance (ability to perform 25% of eating tasks or reliance on parenteral or gastrostomy feedings) to complete independence in feeding. It is more likely that recovery in the total FIM represents recovery over several domains. A one-point improvement in the FIM across five items would represent important changes in a person s quality of life. For example, recovery of eating capability could be from a level of total assistance to one of maximum assistance (ability to perform 25 49% of eating tasks) with similar improvement in four other domains. This degree of improvement in the FIM is in accordance with the modest recovery in motor function scores demonstrated in NASCIS II and in the small number of patients who improved from one large category of motor function to another (for example, from quadriplegic to quadriparetic status). 5,6 These data support the notion that an overall 5.1 improvement in the neurological motor score, which resulted from administration of MP within 8 hours of injury in NASCIS II, 6 would be associated with clinically mean- 264 J. Neurosurg: Spine / Volume 96 / April, 2002

7 Neurological and functional status after SCI TABLE 4 Distribution of FIM improvements by initial injury status, classification of neurological status at 1 year, and grouped expanded score Estimated FIM Improvement (%) No. of Injury Status Patients initial classification of injury complete incomplete neurological status quadriplegic paraplegic quadriparetic paraparetic FIG. 3. Estimated distribution of change in the FIM scores produced by the estimated effect of MP in patients with complete (upper) and incomplete (lower) in NASCIS II. White denotes all available placebo-treated patients; black (overlay), placebo-treated patients with 1-year scores less than 70. ingful improvement to the FIM had it been measured. Of course, 5.1 was the mean motor improvement score with a 95% confidence interval of 0.5 to 9.9, and no measurable MP-induced benefit on their FIM scores would have been demonstrated in some whereas in others greater benefit would be apparent. Similar variability occurs in patients with complete and incomplete injuries. It is neither possible nor necessary to identify these groups of patients further. The results of a randomized trial only confer a likelihood of benefit on an individual patient; they never guarantee it. Clinical trials always undertake an investigation of risk and benefit, and the possibility of complications resulting from high-dose MP must be considered. There has been some comment since the publication of NASCIS II that high-dose steroid therapies pose substantial risk to the patient. These comments are speculative, having been based on historical controlled or uncontrolled small series of patients, 12,13 pharmacological kinetics, 22 and results obtained in animal studies, 24 and they did not present new data. 15,18,19 A systematic review and metaanalysis of the SCI trials does not provide any support for the notion that high-dose MP substantially increases risk for mortality or major morbidity. 3 In a long-term follow-up study of patients investigators failed to find any evidence of avascular J. Neurosurg: Spine / Volume 96 / April, 2002 necrosis. 26 Sauerland, et al., 21 in another recent systematic review and metaanalysis of 51 randomized surgical trials in which high-dose MP treatment was compared with placebo or nothing, provided further reassurance of safety. These trials involved elective, trauma, and spine surgery, and the authors concluded that the surgical procedures were of comparable severity and risk; and their analysis was of steroid- not surgery-related complications. They found no evidence of any increased risk of gastrointestinal bleeding, wound complication, pulmonary complication (for which MP was significantly protective), or death. The numbers of patients who need to be treated to result in one case of harm were 333 for gastrointestinal bleeding, 100 for wound complication, and 29 for pulmonary complications. In discussing these results the authors conclude,...some nonrandomized studies have described serious complications after glucocorticoid administration, such as pneumonia. However, these findings can mainly be explained by the selection of more severely ill patients into an MPSS [methylprednisolone] treatment regimen. 21 Whatever improvement in functional activity is associated with high-dose MP, the most reliable current evidence from properly randomized trials suggests that it is not accompanied by significantly increased risk of adverse effects. Statistical Appendix The outcome compared among the treatment arms of each NASCIS trial was established in the protocol as the change in the overall ASIA score from that obtained at baseline. In addition, the NASCIS II protocol required a comparison of efficacy for early and late initiation of the treatment protocol, which although not explicitly defined in the original proposal, was implemented using a whole number near the median time observed. Therefore, results were reported by whether the patient received the initial bolus within 8 hours of injury. No other time window was analyzed. The study protocol also required an analysis in which patients were stratified by whether neurological injury was diagnosed as complete or incomplete in the emergency room. This was accomplished using the PROC GLM procedure in SAS, which allowed for the fact that this was an unbalanced design. 20 Regression diagnostics revealed that random error had a peculiar distribution that was not normal and that the variance was not homogeneous, thus raising a concern that the variances produced by the software were incorrect. A bootstrap technique was used to validate the estimated standard errors; 11 however, no substantive differences were found. The results from the original analysis of covariance are reported because these are likely to be most easily understood by the widest audience. 265

8 M. B. Bracken and T. R. Holford The FIM scores were not obtained in the patients in NASCIS II, but this was added as an additional measure of treatment efficacy in NASCIS III. We noted in the body of the text, and it can be seen in Fig. 1 that although there is clearly a strong correlation between the overall ASIA score and the FIM, the form for that relationship is not linear. Additionally, because it appears not to follow the shape of a low-order polynomial, nonparametric regression methods were used to describe the relationship quantitatively between the expanded motor score and FIM. The LOESS method used to quantify the relationship between FIM and motor score has the advantage of being able to find an optimally fitting line, without specifying the mathematical form for that line. Thus, the shape of the line is essentially determined by the data. The LOESS method uses locally weighted regression for the values on the x axis, which was the motor score in this study. Weights are defined in a way that results in a fitted FIM, ŷ(x), at a given motor score, x, which depends primarily on the motor scores that are close to it. The weighting is also designed to provide a robust estimate of the fitted line that is, the fitted line avoids being overly influenced by outlying observations. The method was implemented using SAS/INSIGHT. 9 The fitted line showing the estimated relationship between FIM and motor score at one year after SCI is displayed in Fig. 1. To estimate the change in the FIM that would be expected from a specified improvement in the motor score, let x represent treatment-induced change. If x is the 1-year motor score for a placebotreated patient, then one would expect the observed improvement in FIM to be described by the following: ŷ(x) = [ŷ(x x) ŷ(x)]. Figure 2 provides estimates of the change in FIM produced by the overall mean estimate of the effect of MP ( x = 5.1), as well as the estimated effects in patients with complete ( x = 3.6) and incomplete ( x = 7.3) injuries. Also shown in Fig. 2 is the distribution of 1-year motor scores for placebo-treated patients with complete and incomplete injuries. In some of those with incomplete injuries, scores at the maximum were demonstrated and no further treatmentinduced benefit for this measure could have been expected. Disclosure Dr. Bracken has served as an occasional paid consultant to Pharmacia Inc., one of the manufacturers of methylprednisolone. References 1. American Spinal Injury Association, International Medical Society of Paraplegia: International Standards for Neurologic and Functional Classification of Spinal Cord Injury. Chicago: American Spinal Injury Association, Bednarczyk JH, Sanderson DJ: Comparison of functional and medical assessment in the classification of persons with spinal cord injury. J Rehab Res Devel 30: , Bracken MB: Pharmacologic interventions for acute spinal cord injury (Cochrane Review), in: The Cochrane Library Issue 3, Oxford: Update Software 4. Bracken MB, Hildreth N, Freeman DH Jr, et al: Relationship between neurological and functional status after acute spinal cord injury: an epidemiological study. J Chronic Dis 33: , Bracken MB, Shepard MJ, Collins WF, et al: Methylprednisolone or naloxone treatment after acute spinal cord injury: 1-year follow-up data. Results of the second National Acute Spinal Cord Injury Study. J Neurosurg 76:23 31, Bracken MB, Shepard MJ, Collins WF, et al: A randomized, controlled trial of MP or naloxone in the treatment of acute spinal-cord injury. Results of the Second National Acute Spinal Cord Injury Study. New Engl J Med 322: , Bracken MB, Shepard MJ, Holford TR, et al: Administration of methylprednisolone for 24 or 48 hours or tirilazad mesylate for 48 hours in the treatment of acute spinal cord injury. Results of the Third National Acute Spinal Cord Injury Randomized Controlled Trial. National Acute Spinal Cord Injury Study. JAMA 277: , Bracken MB, Shepard MJ, Holford TR, et al: Methylprednisolone or tirilazad mesylate administration after acute spinal cord injury: 1 year follow-up. Results of the third National Acute Spinal Cord Injury Randomized Controlled Trial. J Neurosurg 89: , Cleveland WS: The Elements of Graphing Data. Monterey, CA: Wadsworth Press, Ditunno JF Jr, Graziani V, Tessler A: Neurological assessment in spinal cord injury. Adv Neurol 72: , Efron B, Tibshirani RJ: An Introduction to the Bootstrap. New York: Chapman & Hall, Galandiuk S, Raque G, Appel S, et al: The two-edged sword of large-dose steroids for spinal cord trauma. Ann Surg 218: , Gerndt SJ, Rodriguez JL, Pawlick JW, et al: Consequences of high-dose steroid therapy for acute spinal cord injury. J Trauma 42: , Holford TR, Bracken MB: A model for estimating level and net severity of spinal cord injuries. Stat Med 11: , Hurlbert RJ: Methylprednisolone for acute spinal cord injury: an inappropriate standard of care. J Neurosurg (Spine 1) 93: 1 7, Lazar RB, Yarkony GM, Ortolando D, et al: Prediction of functional outcome by motor capability after spinal cord injury. Arch Phys Med Rehabil 70: , Maynard FM, Bracken MB, Creasey D, et al: International standards for neurological and functional classification of spinal cord injury. American Spinal Injury Association. Spinal Cord 35: , Nesathurai S: Steroids and spinal cord injury: revisiting the NASCIS 2 and NASCIS 3 trials. J Trauma 45: , Qian T, Campagnolo D, Kirshblum S: High-dose methylprednisolone may do more harm for spinal cord injury. Med Hypotheses 55: , SAS Institute Inc., SAS/STAT: User s Guide, Vol. 2, Vers. 6, Ed. 4. Cary, NC: SAS Institute Inc., Sauerland S, Nagelschmidt M, Mallmann P, et al: Risks and benefits of preoperative high dose methylprednisolone in surgical patients: a systematic review. Drug Safe 23: , Segal JL, Maltby BF, Langdorf MI, et al: Methylprednisolone disposition kinetics in patients with acute spinal cord injury. Pharmacotherapy 18:16 22, Shepard MJ, Bracken MB: Magnetic resonance imaging and neurological recovery in acute spinal cord injury: observations from the National Acute Spinal Cord Injury Study 3. Spinal Cord 37: , Short DJ, El Masry WS, Jones PW: High dose methylprednisolone in the management of acute spinal cord injury a systematic review from a clinical perspective. Spinal Cord 38: , Uniform Data System for Medical Rehabilitation. Version 4.0. Buffalo: State University of New York, Wing PC, Nance P, Connell DG, et al: Risk of avascular necrosis following short term megadose methylprednisolone treatment. Spinal Cord 36: , 1998 Manuscript received January 12, Accepted in final form November 13, This study was supported by a grant from the National Institute of Neurological Disorders and Stroke (NS 15078). Address reprint requests to: Michael Bracken, Ph.D., Department of Epidemiology and Public Health, 60 College Street, PO Box , New Haven, Connecticut michael.bracken@ yale.edu. 266 J. Neurosurg: Spine / Volume 96 / April, 2002