Characteristics of Non-traumatic Spinal Cord Dysfunction in Canada Using Administrative Health Data

Abstract

Background: There is a paucity of studies using administrative health data to examine the epidemiology, health care utilization, and outcomes for non-traumatic spinal cord dysfunction (NTSCD). Objective: The purpose of this study is to characterize discrete NTSCD cohorts using decision algorithms with Canadian health administrative databases. Method: Data were provided by the Canadian Institute for Health Information that included all acute care hospital, day surgery, ambulatory, and inpatient rehabilitation records of patients with neurological impairment between April 1, 2004 and March 31, 2011. Diagnostic codes for neurological impairment and NTSCD etiology were used to identify cases and classify 3 NTSCD groups (most likely, probable, and possible). Logistic regression identified factors related to inpatient rehabilitation admission within 7 days of discharge among the preferred group. Results: The most likely NTSCD group (n = 6,362) was significantly older and had a greater proportion of women and individuals with cauda equina lesions compared to the other 2 NTSCD groups (probable [n = 2,777] and possible [n = 11,179]; ps < .001). Factors associated with the likelihood of an inpatient rehabilitation admission included being older (odds ratio [OR], 1.01; 95% CI, 1.00–1.01), being female (OR, 1.18; 95% CI, 1.06–1.32), having paraplegia diagnosis compared to cauda equina (OR, 1.24; 95% CI, 1.09–1.41), residing in an urban area compared to a rural area (OR, 1.34; 95% CI, 1.13–1.58), having degenerative etiology compared to other (OR, 1.59; 95% CI, 1.41–1.80), and having an MRI on record compared to not (OR = 1.57; 95% CI, 1.39–1.76). Conclusion: Administrative data allow for ongoing surveillance of a population in a relatively cost-effective manner. Advancing our knowledge of NTSCD epidemiology, health outcomes, and system performance can inform policy and system planning.

There is a growing body of research to suggest that the prevalence of non-traumatic spinal cord dysfunction (NTSCD) is higher than that of traumatic spinal cord injury,^1–4 despite a greater proportion of research to date on the latter group. It is anticipated that the incidence of NTSCD is increasing within the aging population, as common NTSCD etiologies of injury are age-related.^1,5 Over the years, there has been increased research on NTSCD; however, there is variation on the terms used to identify cases and in the optimal methods to define the inclusion criteria.^6–8 Most research refers to NTSCD as synonymous with non-traumatic spinal cord injury, non-traumatic spinal cord disorder, spinal cord lesion, and non-traumatic spinal cord myelopathy.⁹ In the absence of standardization of nomenclature and case inclusion/exclusion criteria, meaningful comparisons within and across countries remain limited.^6–8 Moreover, monitoring the burden of disease, health system performance, and health outcomes will require valid case identification and tracking across the health system.^1,5,6,10

Most of the published literature on NTSCD has used inpatient rehabilitation chart data from single- and multicenter or survey-based data (often recruiting from inpatient rehabilitation).^11–20 There are biases associated with using these types of data for case finding and incidence and prevalence estimation.²¹ For instance, persons with NTSCD may not be referred to inpatient rehabilitation due to referral patterns and health system resources, which could lead to an underestimate of NTSCD as well as erroneous overall demographic and clinical profiles.⁶ Patients may have received treatment outside of specialized rehabilitation facilities due to personal choice, admission criteria, absence of local treatment facilities available for persons with NTSCD, lengthy wait-times, and/or death prior to rehabilitation facility admission.^11,22,23 Survey data use assumptions when extrapolating to overall population risks, as there is an inherent bias toward persons who have less severe injuries,²⁴ higher socioeconomic status,²⁵ and non-terminal diagnoses.²⁴ An approach beyond a single point of care and survey-based data is required to assess a more inclusive sample population for NTSCD.

To address these biases, the use of administrative health data is ideal for population surveillance including the reporting of NTSCD incidence and prevalence. Unfortunately, there is a paucity of studies using administrative health data for NTSCD.⁶ One of the potential reasons is the lack of consensus on the best methodology to identify NTSCD cases, including accepted international classification diagnostic codes. Improving administrative health data methods for NTSCD case identification will assist with efforts related to public education, prevention, planning of resources, quality indicators, and overall health system improvement. Administrative data allow for ongoing surveillance of a population in a relatively cost-effective manner, especially for lower prevalence conditions such as NTSCD.

Therefore, in response to the need for improvement in case identification,^1,2,6,26 this study contributes to improving methodologies using administrative health data by characterizing the discrete NTSCD cohorts identified using decision algorithms.

Methods

The preceding article in this issue by Jaglal et al describes in detail the methods used for this study.²⁷ Briefly, national administrative health data were used from the Canadian Institute for Health Information (CIHI). All publicly funded health care encounters are captured in health administrative databases collected and stored by CIHI. Data sources included the hospital Discharge Abstract Database (DAD), the National Ambulatory Care Reporting System (NACRS), and the National Rehabilitation Reporting System (NRS).²⁸ The DAD contains all hospital discharge records across Canada (excluding Quebec) with data elements including age, sex, postal code, date of admission, date of discharge, and the most responsible and secondary diagnostic codes (based on the International Classification of Diseases and Related Health Problems, 10th revision [ICD-10-CA]). The NACRS database contains information on all visits to the emergency department (ED), and the main data elements are medical diagnoses and corresponding diagnosis types for the ED visit and patient demographics. The NRS includes patient data from adult inpatient rehabilitation facilities and programs across Canada. CIHI's data are routinely checked for data quality, and the data are comprehensive with high standards.²⁸

The data included Canadian records (excluding the province of Quebec) of all patients 18 years and older with a diagnosis of paraplegia (ICD-10-CA G82.2), tetraplegia (ICD-10-CA G82.5), or cauda equina syndrome (ICD-10-CA G83.4) recorded during an acute care hospital or ED visit or with a primary diagnosis of NTSCD during a rehabilitation stay between April 1, 2004 and March 31, 2011. Of those identified cases, any other records for DAD, NACRS, or NRS visits for any reason between April 1, 2004 and March 31, 2011 were also provided.

Group designations

Patients were included in the “most likely NTSCD” group if they had a diagnosis of paraplegia, tetraplegia, or cauda equina anywhere in the record continuum (eg, hospital, ED, or inpatient rehabilitation) and at least one hospital admission with a NTSCD-related diagnosis code (see Jaglal et al²⁷ for a list of codes) recorded as either the most responsible diagnosis or a pre-existing diagnosis.

Patients were included in the “probable NTSCD” group if they had a diagnosis of paraplegia, tetraplegia, or cauda equina anywhere along record history (eg, hospital, ED, or inpatient rehabilitation) and any of their records included a NTSCD code as a secondary diagnosis type or as a post admission diagnosis type during their record history.

Patients were included in “possible NTSCD” group if they had a diagnosis of paraplegia, tetraplegia, or cauda equina in the hospital during their record history but did not have any recorded NTSCD etiology diagnosis codes in their records.

Variables

Records included clinical and demographic characteristics such as age, sex, neurological impairment (ie, tetraplegia/paraplegia/cauda equina), magnetic resonance imaging (MRI) on record, diagnostic type (eg, most responsible, secondary, post admission), and NTSCD etiology codes. The NTSCD etiology was classified in accordance with the injury details included in the Common Data Elements for SCI.²⁹ Some categories were collapsed due to small number of cases. The following categories were used: degenerative disorder, infections, vascular, oncology malignant, oncology benign, oncology uncertain, inflammatory, inflammatory/infection, congenital/genetic, and unspecified/unknown.

Analysis

To examine the clinical profiles for the probable NTSCD group and the possible NTSCD group, the most responsible ICD-10-CA diagnosis codes were determined for their first acute hospital record within the observation window where the NTSCD code was recorded (as secondary diagnosis) or the neurological impairment (G code) was recorded. The codes were aggregated into the ICD-10-CA chapters to simplify comparisons.³⁰ The timing of diagnoses was also examined. In the most likely NTSCD group, the number of cases admitted to inpatient rehabilitation within 7 days of discharge from the first record with a diagnosis of the neurological impairment G code or NTSCD etiology code in acute care was examined.

Descriptive statistics were reported using means (SD), medians (interquartile range [IQR]), and proportions, as appropriate. Analytical comparisons between group characteristics were conducted using independent t tests, one-way analysis of variance (ANOVA) models with Bonferonni post hoc tests, and chi-square tests as appropriate. Using forced entry logistic regression, the likelihood of being classified as most likely NTSCD compared to the other groups (possible and probable) was regressed on age, sex, rurality, neurological impairment, and MRI in acute care. Given that most literature to date has identified cohorts of NTSCD within inpatient rehabilitation and approximately one-third of the overall cases in this sample were identified in inpatient rehabilitation,²⁷ it was also important to examine the characteristics of persons admitted to rehabilitation compared to those who were not admitted. The likelihood of inpatient rehabilitation admission within 7 days of the acute visit (as standard methods in linking the same episode) was regressed on age, sex, rurality, neurological impairment, NTSCD etiology, and MRI in acute care using a forced entry logistic regression.

Results

Most likely NTSCD

Table 1 shows the characteristics of the most likely NTSCD group (n = 6,362), with an average age of 60.4 years (SD = 17.7); there were 3,601 (56.6%) males and 2,762 females (43.4%). More patients had a diagnosis of paraplegia (n = 3,073; 48.3%), followed by cauda equina (n = 2,150; 33.8%) and tetraplegia (n = 1,140; 17.9%). The largest NTSCD etiology codes were the degenerative (n = 2,684; 42.2%) and the unspecified/unknown categories (n = 1,572; 24.7%). With respect to the timing of diagnoses of the codes, both sets of codes occurred on the same hospital record for 2,386 patients (37.5%); NTSCD etiology code preceded the G code for the majority of patients (n = 3,208; 50.4%), and the G code preceded the NTSCD etiology code for a small number of patients (n = 768; 12.1%). There was a median of 60 days (IQR, 12-359) between diagnoses when the NTSCD etiology code preceded the G code, and a median of 178 days (IQR, 22-632) when the G code came first.

Table 1.

Characteristics of the different non-traumatic spinal cord dysfunction (NTSCD) groups identified using different algorithms of Canadian health administrative data (April 1, 2004 to March 31, 2011) ^a

Of the 6,362 cases identified with the most likely NTSCD group, 32.1% (n = 2,040) were admitted to inpatient rehabilitation within the episode of care (ie, 7 days) of the first acute record with a G code or NTSCD etiology code (see Table 2). Individuals who were admitted to inpatient rehabilitation were significantly younger (median age, 61 years; IQR, 48–73) compared to persons not admitted to inpatient rehabilitation (median age, 66 years; IQR, 54–75; p < .001) and more resided within an urban setting (n = 1,678; 82.3%; p < .001).

Table 2.

Characteristics of the most likely non-traumatic spinal cord dysfunction (NTSCD) group by admission to inpatient rehabilitation status within 7 days of acute care, using Canadian health administrative data (April 1, 2004 to March 31, 2011) ^a

Table 3 shows the results of the logistic regression model. Factors associated with the likelihood of an inpatient rehabilitation admission within 7 days of discharge from acute admission included being older (OR, 1.01; 95% CI, 1.00–1.01), being female (OR, 1.18; 95% CI, 1.06–1.32), having a paraplegia diagnosis compared to cauda equina (OR, 1.24; 95% CI, 1.09–1.41), residing in an urban area compared to a rural area (OR, 1.34; 95% CI, 1.13–1.58), having a degenerative etiology compared to other (OR, 1.59; 95% CI, 1.41–1.80), and having an MRI on record of acute admission compared to not (OR, 1.57; 95% CI, 1.39–1.76).

Table 3.

Admission to inpatient rehabilitation regressed on sociodemographics and clinical characteristics using Canadian health administrative data (April 1, 2004 to March 31, 2011)

Probable NTSCD

The average age for the probable NTSCD group (n = 2,777) was 55.4 years (SD = 20.8). Of the 2,777 patients, 1,632 (58.8%) were male and 1,145 were female (41.2%). A larger proportion had a diagnosis of paraplegia (n = 1,331; 47.9%), followed by tetraplegia (n = 862; 31.0%) and cauda equina (n = 584; 21.0%). The majority had a secondary diagnosis NTSCD etiology code (n = 2,512; 90.5%), mostly related to the degenerative category (n = 703; 25.3%), unspecified/unknown (n = 585; 21.1%), and congenital/genetic (n = 661; 23.8%). A substantial proportion had a most responsible diagnosis within Chapter XXI: Factors influencing health status and contact with health services (n = 699; 25.2%), Chapter XIV: Diseases of the genitourinary system (n = 328; 11.8%), Chapter XIX: Injury, poisoning and other causes (n = 246; 8.9%), and Chapter XIII: Diseases of the musculoskeletal system and connective tissue (n = 224; 8.1%).

Possible NTSCD

The average age of the possible NTSCD group (n = 11,179) was 50.8 years (SD = 19.4) with more males (n = 7,268; 65.0%) than females (n = 3,911; 35.0%). Similar to the probable NTSCD group, more of the patients were diagnosed with paraplegia (n = 5,121; 45.8%), followed by tetraplegia (n = 3,737; 33.4%) and cauda equina (n = 2,321; 20.8%). The most responsible diagnoses codes were equally distributed across the ICD-10 chapter categories. The top represented categories included Chapter XIII: Diseases of the musculoskeletal system and connective tissue (n = 1,484; 13.3%), Chapter VI: Diseases of the genitourinary system (n = 1,228; 11.0%), Chapter VI: Diseases of the nervous system (n = 1,172; 10.5%), and Chapter II: Neoplasms (n = 1,042; 9.3%).

Overall comparisons

Individuals in the most likely NTSCD group were slightly older (OR, 1.02; 95% CI, 1.02–1.03), had a greater proportion of females (OR, 1.25, 95% CI, 1.17–1.34), were more likely to have a MRI on record during the acute admission (OR, 3.40; 95% CI, 3.13–3.71), and had diagnoses of cauda equina with fewer diagnoses of tetraplegia (OR, 0.45; 95% CI, 0.41–0.49) compared to both the probable and possible NTSCD groups. The most likely NTSCD group had a higher proportion of all the NTSCD etiology categories except for metabolic and congenital compared to the probable NTSCD group (p < .001). There were no statistically significant differences between the most likely NTSCD and probable NTSCD groups in the proportion of persons with oncology malignant diagnoses (p > .05), and both groups had a relatively low representation of oncology. Unspecified/unknown diagnoses and degenerative diagnoses were most prevalent for both the most likely NTSCD and probable NTSCD groups.

Discussion

To our knowledge, this is the first Canadian study to use national administrative health data for identifying and characterizing NTSCD patients. Using decision algorithms, we identified 3 potential groups of NTSCD. The characteristics in each group varied substantially depending on whether the most responsible diagnosis or secondary diagnoses were included. These findings highlight the importance of validating administrative data inclusion methodology with patient chart reviews and establishing ongoing international expert panels for agreement on nomenclature, coding, and case inclusion/exclusion criteria.

Our methods are most comparable to work conducted by New in 2008, as administrative data was used to examine the incidence of NTSCD in Victoria, Australia.¹ New used the neurological impairment G codes to identify cases within the acute setting, rather than etiology codes to identify the NTSCD cases. In that work, persons 15 years and older were included, compared to our 18 years and older criteria. New identified an incidence of 26.3 cases per million per year, with rates being higher among males with increasing age.¹ In using the most likely NTSCD group, our incidence was slightly higher at approximately 33.2 cases per million per year (estimates using the Canadian Census 2011 of 27,401,648 Canadians 18+ years of age excluding Quebec).³¹ However, meaningful comparisons of incidence are challenging due to the variety of methodologies used between studies (eg, ICD-10-CA codes and age inclusion criteria). Our original data extraction from CIHI also used neurological impairment (ie, G codes) to identify cases,²⁷ which is similar to New's work¹; however the data required more filtering with the NTSCD etiology codes for enhanced specificity.

In our most likely NTSCD group, we identified slightly more males compared to females and a median age of 63 years. Among those admitted to inpatient rehabilitation (n = 2,040; 32.1%), the median age was 61 years compared to 66 years for those not admitted. Our results reflect similar age and sex distribution among patients with NTSCD within Dutch and Flemish inpatient rehabilitation centers.³² A recent study by Brinkhof identified similar sex distribution but a lower median age of 50 years (IQR, 32–63).²⁴ The variation in age may be due to differences in data sources, as Brinkhof et al used a community survey, where participants were recruited from rehabilitation facilities in Switzerland. While we only captured individuals admitted to inpatient rehabilitation within 7 days of discharge from the acute admission, there were significant differences in characteristics of those admitted to inpatient rehabilitation compared to those who were not, such as location of residence (urban vs rural), sex, type of NTSCD, and neurological impairment. These differences suggest that utilizing data exclusively from rehabilitation facilities could be biased and that a broader approach for capturing an accurate description of the NTSCD is required.

With respect to main etiologies in our most likely NTSCD group, most were related to degenerative causes (n = 2,684; 42.2%) and the unspecified/unknown causes (n = 1,572; 24.7%). The large proportion of degenerative causes is consistent with previous clinical impressions and findings among developed countries.^{6,7,13,20,32,33} However, we did not identify many patients with neoplastic (either benign or malignant) causes contrary to a few published studies.^{5–7,16,20,32,34} This difference may be in part due to the timing of diagnosis as the patient moves from an acute to rehabilitation setting. Unexpectedly, we had a large proportion of unspecified/unknown cases, which may have included some neoplasm cases not yet diagnosed. In comparison, New et al's recent study highlighting characteristics of persons admitted to inpatient rehabilitation with NTSCD across 9 countries (Australia, Canada, Italy, India, Ireland, The Netherlands, Switzerland, United Kingdom, United States) identified 30.8% degenerative causes, 24.9% neoplastic (malignant and benign), and only 1.4% unknown/unspecified cases.³⁴ Considering impairment, if we exclude our cauda equina cases from the most likely NTSCD group, we identified that 72.9% had paraplegia and 27.1% had tetraplegia, which is similar to data reported by New et al with 72.3% of the cases having paraplegia.³⁴

The large proportion of cases with unknown/unspecified etiologies reflects one of the key challenges in using administrative health data for case identification in NTSCD, as the onset date of the etiology, presentation of symptoms, and confirmed diagnoses are not necessarily apparent within a patient's administrative record. As New described in 2014, the temporal variations of symptom presentation may range from minutes (eg, cord infarction) to hours (eg, transverse myelitis), days (eg, spinal abscess), or up to months (eg, spinal canal stenosis).⁸ Our findings highlighted the substantial length in time within a patient's record between the NTSCD etiology code and a G code. There was a median of 60 days (IQR, 12-359) between diagnoses when the NTSCD etiology code preceded the G code, and a median of 178 days (IQR, 22-632) when the G code came first. Only slightly more than one-third of our cases had G codes with the NTSCD etiology codes on the same hospital record; the NTSCD etiology code preceded the G code for the majority of patients (n = 3,208; 50.4%). Therefore, if only neurological impairment G codes are used to identify incident cases, there is a possibility that cases will be captured with a substantial delay in time from the NTSCD diagnosis.

Our study had a few limitations. Our algorithm required patients to be hospitalized in the acute setting; if a person had an NTSCD but was not hospitalized within our observation window, then this person would not be captured with the administrative health data. However, we did explore different care settings such as the ED and inpatient rehabilitation along with a patient's record history to identify where the NTSCD etiology or G code occurred. We did not have an extensive look back period to wash out any chronic NTSCD cases for a true incidence study. In the future, once the NTSCD codes and case identification have been validated, we might plan a longer look back period to ensure only the inclusion of new onset cases; however, we would need to include ICD-9 codes with a longer look back.

Despite these limitations, this work has several strengths. The majority of studies to date on NTSCD utilize data from inpatient rehabilitation and/or recruit participants who have received care from inpatient rehabilitation. There are biases with this type of research in estimating the incidence and prevalence of a population, such as referral biases with admission to inpatient rehabilitation.²¹ In using administrative health data with a single payer health system such as Canada, we are able to capture the overall population at risk with the potential for ongoing surveillance.

The algorithms identified 3 distinct groups with variation in demographics dependent on the decision to select the most responsible diagnosis or secondary diagnoses. Additionally, differences in the patient population admitted to rehabilitation were shown. Our study contributes to a paucity of work on NTSCD using administrative data and supports case identification beyond the admission to rehabilitation. It is important to discuss our approach and findings with international experts to achieve consensus on codes as methodology for case identification.²⁷ Once we have confirmed a validated method, administrative health data can be leveraged to systematically track incidence, prevalence, health care utilization, and system performance for NTSCD. Advancing our knowledge of NTSCD epidemiology, health and well-being outcomes, and health system performance can inform policy and system planning.