Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Jan 1.
Published in final edited form as: Mult Scler Relat Disord. 2021 Oct 24;57:103333. doi: 10.1016/j.msard.2021.103333

Temporal Trends of Multiple Sclerosis Disease Activity: Electronic Health Records Indicators

Liang Liang 1,*, Nicole Kim 1,*, Jue Hou 1, Tianrun Cai 2, Kumar Dahal 2, Chen Lin 3, Sean Finan 3, Guergana Savovoa 3, Mattia Rosso 4, Mariann Polgar-Tucsanyi 4, Howard Weiner 4, Tanuja Chitnis 4, Tianxi Cai 1,5, Zongqi Xia 6
PMCID: PMC8849591  NIHMSID: NIHMS1752830  PMID: 35158446

Abstract

Background:

Long-term data on multiple sclerosis (MS) inflammatory disease activity are limited. We examined electronic health records (EHR) indicators of disease activity in people with MS.

Methods:

We analyzed prospectively collected research registry data and linked EHR data in a clinic-based cohort from 2000 to 2016. We used the trend of the yearly incident relapse rate from the registry data as benchmark. We then calculated the temporal trends of potentially relevant EHR measures, including mean count of the MS diagnostic code, mentions of MS-related concepts, MS-related health utilizations and selected prescriptions.

Results:

1,555 MS patients had both registry and EHR data. Between 2000 and 2016, the registry data showed a declining trend in the yearly incident relapse rate, parallel to an increasing trend of DMT usage. Among the EHR measures, covariate-adjusted frequency of diagnostic code of MS, procedure codes of MS-related imaging studies and emergency room visits, and electronic prescription for steroids declined over time, mirroring the temporal trend of the benchmark yearly incident relapse rate.

Conclusion:

This study highlights EHR indicators of MS relapse that could enable large-scale examination of long-term disease activities or inform individual patient monitoring in clinical settings where EHR data are available.

Keywords: multiple sclerosis, electronic health records, disease activity, relapse

INTRODUCTION

Multiple sclerosis (MS) is a chronic neurological disease (Filippi et al., 2018; Reich et al., 2018; Thompson et al., 2018b) with a high socioeconomic burden (Adelman et al., 2013; Ke et al., 2016). Since 1993, injectable, oral and infusion disease-modifying therapies (DMTs) have been approved based on efficacy in reducing inflammatory disease activity (Baecher-Allan et al., 2018; Li et al., 2020; Rae-Grant et al., 2018). The current approach of MS care emphasizes early initiation and long-term treatment with DMTs to mitigate inflammatory disease activity and delay disability worsening (Corboy et al., 2018; Giovannoni et al., 2016; Kalincik et al., 2021), while research efforts towards stratified or individualized DMT selections are emerging (Bargiela et al., 2017; Kalincik et al., 2017). Studies of the long-term pattern of MS disease activity can inform broader shift in clinical practice.

Electronic health records (EHR) data provide not only real-world evidence of long-term disease trends in chronic diseases but also information for individualized monitoring and prediction of disease activity, when coupled with the appropriate analytical approaches (Liao et al., 2015; Zhang et al., 2019). Prior EHR data research studies in MS, including our own, have accurately classified MS diagnosis (Damotte et al., 2018; Davis et al., 2013; Krysko et al., 2015; Le et al., 2019; Xia et al., 2013). Central to effective EHR data research is the linkage to research registries of well-characterized cohorts (e.g., the Comprehensive Longitudinal Investigation of Multiple Sclerosis at Brigham and Women’s Hospital, CLIMB) as the gold-standard truth (Bove et al., 2017; Gauthier et al., 2006). We previously leveraged EHR and CLIMB data to phenotype MS disease severity (Xia et al., 2013), identify comorbidity burden (Zhang et al., 2020), predict inflammatory disease activity (Ahuja et al., 2021), and conduct comparative effectiveness analysis (Hou et al., 2021). Here, we examined the long-term trend of relapse in the CLIMB participants using registry and EHR indicators between 2000 and 2016.

METHODS

Data Source

The CLIMB study is an ongoing clinic-based long-term prospective study of MS that has been enrolling participants (neurologist-confirmed MS diagnosis, ≥18 years of age, and willing to complete MRI, questionnaire and sample donation yearly) since 2000. The current study used the CLIMB registry data from 2000 to 2016. CLIMB participants have annual clinic visits at a minimum. We previously linked the EHR data for CLIMB participants (Xia et al., 2013) and refreshed the data to include the entire study period. The Mass General Brigham (MGB, formerly Partners HealthCare) IRB approved the study. CLIMB participants provided written consent.

Follow-up Period

Each participant’s follow-up period was the time between the date of the occurrence of the first of any ICD code in the EHR data (or CLIMB cohort enrollment date, whichever occurred earlier) and the date of the occurrence of the last ICD code in the EHR data (or the last visit date in the CLIMB registry, whichever occurred later) between January 1, 2000 and December 31, 2016. We excluded participant data with missing or incomplete years, while we estimated data elements with missing months or days using July 1st or the 15th day of a month, respectively.

Relapse and DMT Data

The CLIMB registry provided: (1) date of clinical and/or radiological relapse; (2) start and end date of DMTs. Clinical relapse was defined as having new or recurrence of neurological symptoms lasting persistently ≥24 hours without fever or infection. Radiological relapse was defined as having either new T1-enhancing lesion and/or new or enlarging T2-FLAIR lesion on brain, orbit or spinal cord magnetic resonance imaging (MRI) studies based on clinical radiology reports. DMTs included injectable [interferon-beta (all brands), glatiramer acetate], oral [dimethyl fumarate, fingolimod, teriflunomide], and infusion [natalizumab, rituximab]. We excluded several DMTs from analyses due to small sample size (i.e., alemtuzumab, daclizumab, mitoxantrone) or approval date after 2016 (e.g., ocrelizumab, siponimod).

EHR Data

The MGB EHR provided longitudinally recorded codified and narrative variables from hospitalizations and outpatient visits (Xia et al., 2013). Codified data included all International Classification of Disease (ICD) codes, selected Current Procedural Terminology (CPT) codes (i.e., MRI of brain, orbit and spine; hospital admissions; emergency department visits) and selected electronic prescriptions (e.g., methylprednisolone). Given that ICD-10 was implemented only during the final years of the study period, the majority of the ICD data were ICD-19. As such, we mapped all ICD-10 codes to ICD-9 codes (Services, n.d.). We deployed the Apache clinical Text Analysis and Knowledge Extraction System (cTAKES: ctakes.apache.org) for natural language processing (NLP) on clinical narratives (e.g., encounter notes, discharge summaries, radiology reports) to generate concept unique identifiers (CUIs) for clinical concepts (Savova et al., 2010). Our analyses included positive mentions of CUIs after excluding negation (e.g., “no evidence of”), family history, conditional (if, then) or temporality (future tense).

Using the EHR data, we computed the counts of potential indicators of relapse based on ICD and CPT codes, CUI mentions and pertinent electronic prescriptions among CLIMB participants. To avoid duplicate counts, we counted ICD and CPT codes as the number of days for a given code, i.e., no more than one code per day. For MS-related CPT codes, we grouped related CPT codes according to the American Medical Association guideline (Supplementary Table 1). For narrative mentions of “multiple sclerosis”, “relapse”, and “DMT”, we grouped related CUIs of each category of mentions (Supplementary Table 2). To avoid inflating narrative mentions, we took the highly conservative approach of counting multiple occurrences of the same category of narrative mention of “multiple sclerosis”, “relapse”, and “DMT” in each given year once. For selected electronic prescriptions, we grouped prescription codes in the same category of medication (Supplementary Table 3).

Temporal Trends

Using the CLIMB registry data, we computed: (1) the yearly incident relapse rate as the total number of relapses (clinical and/or radiological) over the total number of CLIMB participants in each given year between 2000 and 2016; (2) the yearly DMT usage as the percentage of participants receiving any of the approved DMTs (plus rituximab) in a given year. We defined disease duration for each participant in a given year as the time period between the first symptom date and December 31st of that year or the last date of patient enrollment in the study, whichever came first. For all temporal trends, we calculated the average rate of change per year by fitting a linear regression of a yearly rate over year and reported the coefficient and two-sided p-value for linear trend over the study period and the 95% confidence intervals estimated using the bootstrap method.

To compute the temporal trends using potential EHR indicators of relapse, we aggregated all patient-level data by year and analyzed the longitudinal observations over years using the generalized estimating equation approach to account for within-patient correlation. We used Poisson regression for count variables (e.g., ICD, CPT and prescription counts) and logistic regression for binary variables (e.g., the presence of CUIs). Importantly, we used covariate adjusted means in the temporal trend analyses to eliminate potential bias due to the change in patient population over time. Specifically, the analyses adjusted for healthcare utilization (defined as the total number of notes and the total number of all ICD codes in the EHR system) and key patient characteristics (age at symptom onset, race / ethnicity, sex and disease duration). The number of notes and the number of ICD codes were transformed via x->log(1+x) to normalize distributions. We performed analyses using R (3.4.2).

Code and Data Availability

Code for analysis and figure generation is available at <https://github.com/nkim1322/MS_EHR_paper1>. Summary data that support the findings of this study as well as anonymous data used for this study are available upon reasonable request to the corresponding author.

RESULTS

Patient Characteristics

Between 2000 and 2016, 1,555 MS patients who enrolled in the CLIMB prospective cohort study had available EHR data (Table 1). As expected for a patient population treated in a MS specialty clinic, annualized relapse rate in this cohort was low: 0.145±0.01. Participants who were ever treated with any DMT had modestly but significantly higher annualized relapse rate than those never treated with a DMT (0.175±0.011 vs. 0.01±0.005, p<0.0001) (Table 1).

Table 1.

Characteristics of the study population and the subgroups based on ever versus never treatment status with disease modifying treatments.

Parameters CLIMB-EHR Cohort Ever Treated Never Treated P-value
Total number of patients 1555 1247 308 NA
Sex, % women1 73.9% 74.0% 73.4% 0.824
Race, % non-Hispanic European descent1 92.6% 92.2% 93.0% 0.733
Median (IQR) age at first ICD code, years2 35.2 (16.2) 34.9 (16.2) 36.8 (15.7) 0.101
Median (IQR) age at symptom onset, years 32.4 (14.3) 32.2 (14.6) 32.8 (13.3) 0.368
Median (IQR) age at CLIMB enrollment, years 32.8 (14.9) 32.5 (15.0) 34.0 (14.6) 0.129
Median (IQR) duration of follow-up, years3 17.1 (11.4) 17.8 (11.1) 14.3 (12.4) <0.0001
Median (IQR) disease duration, years 8.5 (13) NA NA NA
Median (IQR) number of DMT during follow-up4 2 (3) 3 (2) 0 (0) NA
% patients receiving any DMT4 80.2% 100% 0% NA
Median (IQR) EDSS at CLIMB enrollment5 3 (4) NA NA NA
Annualized relapse rate, mean (SD)6 0.145 (0.010) 0.175 (0.011) 0.010 (0.005) <0.0001
Open in a new tab

Note:

1.

We excluded missing values in sex (0.4%) and race (2.9%).

2.

The first of any ICD code in the EHR data.

3.

See Methods for definition of follow-up duration.

4.

See Table 2 for the list of DMT.

5.

EDSS, Expanded Disability Status Scale.

6.

Relapse includes clinical and/or radiological, between 2000 and 2016.

Temporal Trends of Relapse and DMT Usage

We first established the gold-standard temporal trends of inflammatory disease activity in the study population using the CLIMB registry data. The temporal trend of relapse (clinical and/or radiological) between 2000 and 2016 showed a significant decline in the yearly incident relapse rate (average rate of change per year=−7.7×10−3 relapses per patient-year, [95% CI: −9.0×10−3, −6.4×10−3], p-trend<0.0001) (Figure 1). The declining trend in the yearly incident relapse rate was driven primarily by the yearly incident clinical relapse rate (−6.0×10−3 relapses per patient-year, [95% CI: −7.0×10−3, −4.9×10−3], p-trend<0.0001), and to a lesser extent by the yearly incident radiological relapse rate (−1.4×10−3 relapses per patient-year, [95% CI: −1.9×10−3, −9.7×10−4], p-trend<0.0001) (Supplementary Results 1, Supplementary Figure 1).

Figure 1.

Figure 1.

Open in a new tab

Temporal trends of yearly incident relapse rate (clinical and/or radiological), percentage of total participants receiving any DMT, yearly incident relapse rate among patients receiving any DMT versus untreated patients in a given year in the CLIMB cohort between 2000 and 2016. 95% confidence intervals are shown in shade. Two-sided p-value for linear trend indicates the significance level of the overall trend across the entire study period. “N” at the top of the figure indicates the number of CLIMB participants in a given year.

Between 2000 and 2016, there was a significant increase in the percentage of CLIMB participants receiving any DMT (average rate of change=2.7% per year, [95% CI: 2.4%–2.9%], p-trend<0.0001) (Figure 1). Interferon-beta and glatiramer acetate were the most frequently prescribed DMT classes and had the longest treatment duration, reflecting the earlier approval dates of these DMTs (Table 2). The usage of injectable drugs (interferon-beta, glatiramer acetate) declined over the study period, coinciding with increasing usage of oral (dimethyl fumarate, fingolimod, and teriflunomide) and infusion drug (natalizumab, rituximab) (Figure 2), consistent with known clinical practice. The usage of high-efficacy DMTs (natalizumab or rituximab: rate=1.1% per year, [95% CI: 0.93%–1.12%], p-trend<0.0001) and standard-efficacy DMTs (dimethyl fumarate, fingolimod, glatiramer acetate, interferon-beta, and teriflunomide: rate=1.8% per year, [95% CI: 1.5%–2.1%], p-trend<0.0001) both increased (Supplementary Figure 2).

Table 2.

Prescription pattern of disease-modifying treatments (DMT) in the study cohort.

DMT Number of Patients Treatment Duration Months, Median (IQR) Inclusion in Analysis2 (Y/N)
No Treatment 308 NA Y
Alemtuzumab 1 13.67 (0) N
Daclizumab 65 15.6 (28.8) N
Dimethyl fumarate 298 26.9 (33.6) Y
Fingolimod 288 30.5 (35.5) Y
Glatiramer acetate 797 42.0 (50.3) Y
Interferon-beta1 1095 61.7 (94.6) Y
Mitoxantrone 19 58.7 (96.0) N
Natalizumab 265 29.7 (55.7) Y
Rituximab 192 24.5 (55.7) Y
Teriflunomide 87 16.5 (25.9) Y
Open in a new tab

Note:

1.

Interferon-beta includes IFN-beta-1a, IFN-beta-1b, IFN-beta-1a-PEG.

2.

The following DMTs were from analyses due to the small subgroup size in this data set during the study period (mitoxantrone, alemtuzumab, and daclizumab) or approval after the end of the study period (e.g., ocrelizumab, siponimod).

Figure 2.

Figure 2.

Open in a new tab

Temporal trends of the percentage of CLIMB participants receiving any of the disease-modifying treatments between 2000 and 2016. 95% confidence intervals are shown in shade.

Temporal Trend of Potential EHR Indicators of Relapse

To examine whether EHR data can inform long-term patterns of disease activity, we investigated the temporal trends of potential EHR-derived indicators of relapse in CLIMB participants. Given that only one MS diagnostic code exists in either the ICD-9 or ICD-10 system and that there is no specific code indicating MS relapse, we first examined the temporal trend of the frequency of MS ICD code as a disease activity proxy (Figure 3). There was an increase in the frequency of MS ICD code (average rate of change=0.15 counts per patient-year, [95% CI: 0.12–0.18], p-trend<0.0001), parallel to the increase in the frequency of any ICD code (rate=0.71 counts per patient-year, [95% CI: 0.62, 0.81], p-trend<0.0001). After accounting for healthcare utilization and patient characteristics, the adjusted frequency of MS ICD code per year indeed declined during the study period (rate= −9.8×10−2 counts per patient-year, [95% CI: −1.1×10−1, −8.4×10−2], p-trend<0.0001), mirroring the declining trend of yearly incident relapse rate in the CLIMB registry data, which served as the benchmark.

Figure 3.

Figure 3.

Open in a new tab

Temporal trends of International Classification of Disease (ICD) code in CLIMB participants between 2000 and 2016: mean count of ICD code for MS, mean count of any ICD code, and adjusted mean count of ICD code for MS. We adjusted for healthcare utilization (the number of notes and the number of any ICD code in the EHR system) and key patient characteristics (age at symptom onset, race, ethnicity, sex and disease duration). 95% confidence intervals are shown in shade. Two-sided p-value for linear trend indicates the significance level of the overall trend across the entire study period. “N” indicates the number of CLIMB participants with EHR data. “n” indicates the number of participants with ICD code for MS in the EHR.

We next examined the temporal trend of narrative mentions of MS-relevant concepts (Supplementary Table 2, Figure 4). In covariates-adjusted analysis, there was an increasing trend in the adjusted frequency of narrative mention of “multiple sclerosis” (average rate of change=2.2×10−3 mentions per patient-year, [95% CI: 2.1×10−3, 2.4×10−3], p-trend<0.0001). Similarly, the trend of the adjusted frequency of mention of “relapse” increased (rate=1.4×10−2 mentions per patient-year, [95% CI: 1.4×10−2, 1.4×10−2], p-trend<0.0001). The increasing rates of narrative mentions of “multiple sclerosis” and “relapse” in the EHR data contrasted with the declining trends of the adjusted frequency of the MS ICD code and the benchmark incident relapse rate from the registry data. On the other hand, the increasing rate in the narrative mention of any “DMT” (rate=9.7×10−3 mentions per patient-year, [95% CI: 9.4×10−3, 1.0×10−2], p-trend<0.0001) was consistent with the increasing trend of DMT usage in the registry data. When examining only encounter notes that also contain mentions of “multiple sclerosis”, we observed the same increasing rates of the adjusted narrative mentions of “relapse” (rate=1.6×10−2 mentions per patient-year, [95% CI: 1.6×10−2, 1.6×10−2], p-trend<0.0001) and “DMT” (rate=1.2×10−2 mentions per patient-year, [95% CI: 1.2×10−2, 1.2×10−2], p-trend<0.0001).

Figure 4.

Figure 4.

Open in a new tab

Temporal trends of MS-related mentions (or Concept Unique Identifiers, CUIs) in the clinical narratives of CLIMB participants between 2000 and 2016: adjusted mean count of mentions of “multiple sclerosis”, adjusted mean count of mentions of “relapse”, and adjusted mean count of mentions of any of the “disease-modifying treatments, DMTs”. Please refer to Supplementary Table 2 for description of the CUIs in each category of MS-related CUIs. We adjusted for healthcare utilization (the number of notes and the number of any ICD code in the EHR system) and key patient characteristics (age at symptom onset, race, ethnicity, sex and disease duration). 95% confidence intervals are shown in shade. Two-sided p-value for linear trend indicates the significance level of the overall trend across the entire study period. “N” indicates the number of CLIMB participants with EHR data. “n” indicates the number of patients with the specific category of narrative data (CUIs) in the EHR data. The red lines indicate the adjusted mean count of narrative mentions of “relapse” and “DMT” that co-exist in the same notes as mentions of “multiple sclerosis”.

We then examined the temporal trend of MS-related procedure codes (Supplementary Table 1, Figure 5). In covariate-adjusted analysis, the temporal trend of the adjusted frequency of MRI procedures (brain, orbit or spinal cord) (average rate of change= −2.6×10−2 procedures per patient-year, [95% CI: −2.8×10−2, −2.3×10−2], p-trend<0.0001) and emergency department visits (rate= −8.9×10−3 visits per patient-year, [95% CI: −1.0×10−2, −7.5×10−3], p-trend<0.0001) both significantly declined over time. There was no significant change in the adjusted rate of hospital admissions during the study period (p-trend=0.45). Neither emergency department visit nor hospital admission was specific for MS as the primary reason. For reference, the proportion of participants with any MRI procedure (brain, orbit, or spinal cord) (rate=2.0×10−2 procedures per patient-year, [95% CI: 1.7×10−2, 2.2×10−2], p-trend<0.0001), the proportion of participants with any emergency department visits (rate=1.9×10−3 visits per patient-year, [95% CI: 1.2×10−3, 2.7×10−3], p-trend<0.0001) and the proportion of participants with any hospital admissions (rate= 2.7×10−3 admissions per patient-year, [95% CI: 2.1×10−3, 3.3×10−3], p-trend<0.0001) all showed increasing temporal trends.

Figure 5.

Figure 5.

Open in a new tab

Temporal trends of MS-related health utilization (Current Procedural Terminology or CPT codes) in CLIMB participants between 2000 and 2016: adjusted mean number of MRI scans (brain, orbit, or spinal cord, person-year), adjusted mean number of hospital admissions (person-year), adjusted mean number of emergency department visits (person-year). Please refer to Supplementary Table 1 for description of the CPT codes in each category. We adjusted for healthcare utilization (the number of notes and the number of any ICD code in the EHR system) and key patient characteristics (age at symptom onset, race, ethnicity, sex and disease duration). 95% confidence intervals are shown in shade. Two-sided p-value for linear trend indicates the significance level of the overall trend across the entire study period. “N” indicates the number of participants in the CLIMB cohort with linked EHR data. “n” indicates the number of patients with the specific category of health utilization data (CPT codes) in the EHR. The red lines indicate the proportion of participants with any MRI scans of brain, orbit or spinal cord, any hospital admissions or any emergency department visits over time.

Finally, we examined the temporal trend of electronic prescriptions of steroid (including inpatient, outpatient and home infusion as well as oral formulation of methylprednisolone, Supplementary Table 3) between 2006 and 2016. High-dose steroid such as methylprednisolone is frequently prescribed for acute relapse. We focused on this shorter time interval than the overall study period, because the MGB Healthcare system fully implemented electronic prescriptions in 2006. We observed a significant decline in the adjusted frequency of steroid prescriptions (rate= −1.4×10−1 prescriptions per patient-year, [95% CI: −1.5×10−1, −1.3×10−1], p-trend<0.0001) (Supplementary Figure 3). For comparison, the adjusted frequency of prescriptions of the oldest category of DMTs, interferon-beta, showed a declining trend over the same study period (rate= −1.1×10−2 prescriptions per patient-year, [95% CI: −1.2×10−2, −1.0×10−2], p-trend<0.0001) (Supplementary Figure 3), consistent with the registry finding.

In supplementary analysis of the four EHR indicators of MS relapse whose temporal trends mirrored that of the incident rate of the registry-annotated relapse, we assessed the temporal trends of the ratios of these covariate-adjusted EHR indicators of MS relapse as a function of the observed actual relapse over time to confirm that these findings were not spurious (Supplementary Result 2, Supplementary Figure 4).

DISCUSSION

We examined the long-term temporal trends of inflammatory disease activity using both registry data and EHR indicators in a large well-characterized prospective MS cohort. The benchmark registry data showed a declining yearly incident relapse rate between 2000 and 2016. We identified several EHR indicators that mirrored the benchmark temporal trend of yearly incident relapse rate: adjusted frequency of the MS diagnostic code, procedure codes of MS-related MRI scans and emergency room visits, and steroid prescriptions. These EHR indicators could be informative when evaluating the real-world evidence of disease patterns on a large scale or when guiding individual patient monitoring in clinical settings where EHR data are readily available.

The analysis of the gold-standard registry data showed a long-term decline in MS relapse rate in the CLIMB cohort. The finding of the greater declining incident clinical than radiological relapse rate could be attributable to ascertainment bias: clinical relapses may be better recorded than radiological relapses. The declining trend in incident relapse rate in the untreated patients, though to a lesser extent than the DMT-treated, could reflect the practice of not treating “benign” patients. Indeed, never-treated participants had a lower annualized incidence rate than those who were ever treated with any DMT. We also could not exclude that patients were untreated due to lack of effective DMT, particularly for progressive MS during the study period. Thus, great caution is required when interpreting the yearly incident relapse rate between treated and untreated patients (and between high- and standard-efficacy DMTs) due to indication bias and floor effect (in the untreated). Other contributing factors to the declining trend of incident relapse rate included the changing diagnostic criteria of MS during the study period and the increasing patient age and disease duration as the cohort matures. Changing diagnostic criteria led to earlier diagnosis and interventions, which reduce long-term disease activity (Polman et al., 2005, 2011; Thompson et al., 2018a). Further, relapse frequency declines with age, which is the rationale for an ongoing randomized clinical trial examining the safety of discontinuing DMTs in elderly patients with stable disease activity (clinicaltrials.gov: NCT03073603). Aside from these caveats, the purpose of using the registry data to compute the relapse trend is to provide the benchmark for assessing EHR indicators of relapse.

Our study of the EHR indicators of MS relapse built on our prior work of integrating EHR with research registry data to phenotype multiple sclerosis severity score and brain volume (Xia et al., 2013), systemically identify the comorbidity burden associated with MS severity using a phenome-wide approach (Zhang et al., 2020) and predict future relapse using a clinically deployable tool (Ahuja et al., 2021), and conduct comparative effectiveness analysis (Hou et al., 2021). This integrated approach using the registry data as the gold-standard is a key strength of our EHR method. Here, we found that the temporal trends of the adjusted frequency of the MS diagnostic code, procedure codes of MS-related MRI scans and emergency room visits, and steroid prescription mirrored the temporal trend of the registry-defined benchmark (i.e., yearly incident relapse rate). After accounting for confounders, the temporal trends of mentions of “multiple sclerosis” and “relapse” did not mirror the temporal trend of the benchmark yearly incident relapse rate, suggesting that a more sophisticated modeling of narrative EHR indicators (e.g., counting co-reference categories instead of phrases that refer to the same entities) will be more suitable.

Our study has a limitation of potential generalizability. Given that CLIMB participants receive neurological care at an academic specialty clinic in the United States, the study population has a low average annualized relapse rate, which is comparable to other clinic-based registries. While the findings of the registry data may not be representative of the broader care environment, this data set provides the unique combination of research registry data with linked EHR data to enable the temporal analysis of the EHR indicators of MS disease activity, which could have broader application. Research registry data provide the invaluable gold-standard benchmark for EHR data research. Given the wide adoption of EHR, EHR indicators of relapse developed using the gold-standard benchmark could enable large-scale studies of long-term MS disease activity patterns across diverse clinical settings. This EHR-based approach would complement efforts of physician-centered research registries such as CLIMB or MSBase (Kalincik et al., 2014).

Our study highlights relevant EHR indicators of inflammatory MS disease activity that could enable large-scale studies of long-term disease patterns and potentially inform individual disease monitoring, particularly in settings where EHR data are readily available. Further studies examining the relationship between the dynamic patterns in the EHR data and long-term disease outcomes (relapse and disability) are warranted.

Supplementary Material

Supplementary Material

HIGHLIGHTS.

  • The benchmark yearly incident multiple sclerosis relapse rate declined between 2000 and 2016.

  • We examined the covariate-adjusted electronic health records (EHR) indicators of multiple sclerosis disease activity over the study period.

  • The adjusted frequency of diagnostic code of multiple sclerosis, procedure codes of MS-related magnetic resonance imaging studies and emergency department visits, and electronic prescription for steroids declined over the same period.

  • These EHR indicators of MS relapse could enable large-scale examination of long-term disease activity trends or inform individual patient monitoring.

ACKNOWLEDGMENTS

We thank all Brigham MS Clinic neurologists who contributed to the study through patient examination and data collection as well as all research participants for their time and efforts. We thank Andrew Cagan and Vivian Gainer for providing technical assistance with the EHR data.

Appendix 1. List of authors

Name Location Role Contribution
Liang Liang, PhD Harvard T. H. Chan School of Public Health, Boston Author Designed and conceptualized study; analyzed data; drafted the manuscript for intellectual content.
Nicole Kim, BA Harvard T. H. Chan School of Public Health, Boston Author Designed and conceptualized study; analyzed data; drafted and revised the manuscript for intellectual content.
Jue Hou, PhD Harvard T H. Chan School of Public Health, Boston Author Analyzed data; Interpreted the data; revised manuscript for intellectual content
Tianrun Cai, MD, PhD Brigham and Women’s Hospital, Boston Author Major role in the data acquisition
Kumar Dahal, MS Brigham and Women’s Hospital, Boston Author Major role in the data acquisition
Chen Lin, MS Boston Children’s Hospital, Boston Author Major role in the data acquisition
Sean Finan, BS Boston Children’s Hospital, Boston Author Major role in the data acquisition
Guergana Savova, PhD Boston Children’s Hospital, Boston Author Major role in the data acquisition
Mattia Rosso, MD Brigham and Women’s Hospital, Boston Author Major role in the data acquisition
Mariann Polgar-Tucsanyi, MS Brigham and Women’s Hospital, Boston Author Major role in the data acquisition
Howard Weiner, MD Brigham and Women’s Hospital, Boston Author Major role in the data acquisition
Tanuja Chitnis, MD Brigham and Women’s Hospital, Boston Author Major role in the data acquisition
Tianxi Cai, ScD Harvard T. H. Chan School of Public Health, Boston Author Designed and conceptualized study; analyzed data; revised the manuscript for intellectual content.
Zongqi Xia, MD, PhD University of Pittsburgh, Pittsburgh Author Designed and conceptualized study; analyzed data; drafted and revised the manuscript for intellectual content.
Open in a new tab

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

REFERENCE

  1. Adelman G, Rane SG, Villa KF, 2013. The cost burden of multiple sclerosis in the United States: a systematic review of the literature. Journal of medical economics 16, 639–647. 10.3111/13696998.2013.778268 [DOI] [PubMed] [Google Scholar]
  2. Ahuja Y, Kim N, Liang L, Cai Tianrun, Dahal K, Seyok T, Lin C, Finan S, Liao K, Savovoa G, Chitnis T, Cai Tianxi, Xia Z, 2021. Leveraging electronic health records data to predict multiple sclerosis disease activity. Ann Clin Transl Neur. 10.1002/acn3.51324 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Baecher-Allan C, Kaskow BJ, Weiner HL, 2018. Multiple Sclerosis: Mechanisms and Immunotherapy. Neuron 97, 742–768. 10.1016/j.neuron.2018.01.021 [DOI] [PubMed] [Google Scholar]
  4. Bargiela D, Bianchi MT, Westover MB, Chibnik LB, Healy BC, Jager PLD, Xia Z, 2017. Selection of first-line therapy in multiple sclerosis using risk-benefit decision analysis. Neurology 88, 677–684. 10.1212/wnl.0000000000003612 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Bove R, Chitnis T, Cree BA, Tintoré M, Naegelin Y, Uitdehaag BM, Kappos L, Khoury SJ, Montalban X, Hauser SL, Weiner HL, Consortium S, 2017. SUMMIT (Serially Unified Multicenter Multiple Sclerosis Investigation): creating a repository of deeply phenotyped contemporary multiple sclerosis cohorts. Multiple Sclerosis Journal 1352458517726657. 10.1177/1352458517726657 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Corboy JR, Weinshenker BG, Wingerchuk DM, 2018. Comment on 2018 American Academy of Neurology guidelines on disease-modifying therapies in MS. Neurology 90, 1106–1112. 10.1212/wnl.0000000000005574 [DOI] [PubMed] [Google Scholar]
  7. Damotte V, Lizee A, Tremblay M, Agrawal A, Khankhanian P, Santaniello A, Gomez R, Lincoln R, Tang W, Chen T, Lee N, Villoslada P, Hollenbach JA, Bevan CD, Graves J, Bove R, Goodin DS, Green AJ, Baranzini SE, Cree BA, Henry RG, Hauser SL, Gelfand JM, Gourraud P-A, 2018. Harnessing electronic medical records to advance research on multiple sclerosis. Multiple Sclerosis Journal 135245851774740. 10.1177/1352458517747407 [DOI] [PubMed] [Google Scholar]
  8. Davis MF, Sriram S, Bush WS, Denny JC, Haines JL, 2013. Automated extraction of clinical traits of multiple sclerosis in electronic medical records. Journal of the American Medical Informatics Association 20, e334–40. 10.1136/amiajnl-2013-001999 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Filippi M, Bar-Or A, Piehl F, Preziosa P, Solari A, Vukusic S, Rocca MA, 2018. Multiple sclerosis. Nat. Rev. Dis. Primers 4(43). 10.1038/s41572-018-0041-4 [DOI] [PubMed] [Google Scholar]
  10. Gauthier SA, Glanz BI, Mandel M, Weiner HL, 2006. A model for the comprehensive investigation of a chronic autoimmune disease: the multiple sclerosis CLIMB study. Autoimmunity Reviews 5, 532–536. 10.1016/j.autrev.2006.02.012 [DOI] [PubMed] [Google Scholar]
  11. Giovannoni G, Butzkueven H, Dhib-Jalbut S, Hobart J, Kobelt G, Pepper G, Sormani MP, Thalheim C, Traboulsee A, Vollmer T, 2016. Brain health: time matters in multiple sclerosis. Multiple Sclerosis and Related Disorders 9 Suppl 1, S5–S48. 10.1016/j.msard.2016.07.003 [DOI] [PubMed] [Google Scholar]
  12. Hou J, Kim N, Cai T, Dahal K, Weiner H, Chitnis T, Cai T, Xia Z. Comparison of dimethyl fumarate vs fingolimod and rituximab vs natalizumab for treatment of multiple sclerosis. JAMA Netw Open. 2021;4(11):e2134627. doi:10.1001/jamanetworkopen.2021.34627 https://pubmed.ncbi.nlm.nih.gov/34783826/ [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Kalincik T, Buzzard K, Jokubaitis V, Trojano M, Duquette P, Izquierdo G, Girard M, Lugaresi A, Grammond P, Grand’Maison F, Oreja-Guevara C, Boz C, Hupperts R, Petersen T, Giuliani G, Iuliano G, Lechner-Scott J, Barnett M, Bergamaschi R, Pesch VV, Amato MP, Munster E. van, Fernandez-Bolanos R, Verheul F, Fiol M, Cristiano E, Slee M, Rio ME, Spitaleri D, Alroughani R, Gray O, Saladino ML, Flechter S, Herbert J, Cabrera-Gomez JA, Vella N, Paine M, Shaw C, Moore F, Vucic S, Savino A, Singhal B, Petkovska-Boskova T, Parratt J, Sirbu C-A, Rozsa C, Liew D, Butzkueven H, Group, Msb.S., 2014. Risk of relapse phenotype recurrence in multiple sclerosis. Mult Scler J 20, 1511–1522. 10.1177/1352458514528762 [DOI] [PubMed] [Google Scholar]
  14. Kalincik T, Diouf I, Sharmin S, Malpas C, Spelman T, Horakova D, Havrdova EK, Trojano M, Izquierdo G, Lugaresi A, Prat A, Girard M, Duquette P, Grammond P, Jokubaitis V, Walt A. van der, Grand’Maison F, Sola P, Ferraro D, Shaygannejad V, Alroughani R, Hupperts R, Terzi M, Boz C, Lechner-Scott J, Pucci E, Pesch VV, Granella F, Bergamaschi R, Spitaleri D, Slee M, Vucic S, Ampapa R, McCombe P, Ramo-Tello C, Prevost J, Olascoaga J, Cristiano E, Barnett M, Saladino ML, Sanchez-Menoyo JL, Hodgkinson S, Rozsa C, Hughes S, Moore F, Shaw C, Butler E, Skibina O, Gray O, Kermode A, Csepany T, Singhal B, Shuey N, Piroska I, Taylor B, Simo M, Sirbu C-A, Sas A, Butzkueven H, Group, Msb.S., 2021. Effect of Disease Modifying Therapy on Disability in Relapsing-Remitting Multiple Sclerosis Over 15 Years. Neurology 96, e783–e797. 10.1212/wnl.0000000000011242 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Kalincik T, Manouchehrinia A, Sobisek L, Jokubaitis V, Spelman T, Horakova D, Havrdova E, Trojano M, Izquierdo G, Lugaresi A, Girard M, Prat A, Duquette P, Grammond P, Sola P, Hupperts R, Grand’maison F, Pucci E, Boz C, Alroughani R, Pesch VV, Lechner-Scott J, Terzi M, Bergamaschi R, Iuliano G, Granella F, Spitaleri D, Shaygannejad V, Oreja-Guevara C, Slee M, Ampapa R, Verheul F, McCombe P, Olascoaga J, Amato MP, Vucic S, Hodgkinson S, Ramo-Tello C, Flechter S, Cristiano E, Rozsa C, Moore F, Sanchez-Menoyo JL, Saladino ML, Barnett M, Hillert J, Butzkueven H, Group, Msb.S., 2017. Towards personalized therapy for multiple sclerosis: prediction of individual treatment response. Brain 140, 2426–2443. 10.1093/brain/awx185 [DOI] [PubMed] [Google Scholar]
  16. Ke X, Navaratnam P, Sasane R, Lawrence DFE, Friedman HS, Tulsi BB, Vollmer T, 2016. Determinants of high cost in multiple sclerosis patients: a claims and chart review study. Curr Med Res Opin 32, 1–47. 10.1080/03007995.2016.1192529 [DOI] [PubMed] [Google Scholar]
  17. Krysko KM, Ivers NM, Young J, O’Connor P, Tu K, 2015. Identifying individuals with multiple sclerosis in an electronic medical record. Multiple Sclerosis Journal 21, 217–224. 10.1177/1352458514538334 [DOI] [PubMed] [Google Scholar]
  18. Le HV, Truong CTL, Kamauu AWC, Holmén J, Fillmore C, Kobayashi MG, Martin C, Sabidó M, Wong SL, 2019. Identifying Patients With Relapsing-Remitting Multiple Sclerosis Using Algorithms Applied to US Integrated Delivery Network Healthcare Data. Value Health 22, 77–84. 10.1016/j.jval.2018.06.014 [DOI] [PubMed] [Google Scholar]
  19. Li H, Hu F, Zhang Y, Li K, 2020. Comparative efficacy and acceptability of disease-modifying therapies in patients with relapsing–remitting multiple sclerosis: a systematic review and network meta-analysis. J Neurol 267, 3489–3498. 10.1007/s00415-019-09395-w [DOI] [PubMed] [Google Scholar]
  20. Liao KP, Cai T, Savova GK, Murphy SN, Karlson EW, Ananthakrishnan AN, Gainer VS, Shaw SY, Xia Z, Szolovits P, Churchill S, Kohane I, 2015. Development of phenotype algorithms using electronic medical records and incorporating natural language processing. BMJ 350, h1885. 10.1136/bmj.h1885 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Polman CH, Reingold SC, Banwell B, Clanet M, Cohen JA, Filippi M, Fujihara K, Havrdova E, Hutchinson M, Kappos L, Lublin FD, Montalban X, O’Connor P, Sandberg-Wollheim M, Thompson AJ, Waubant E, Weinshenker B, Wolinsky JS, 2011. Diagnostic criteria for multiple sclerosis: 2010 revisions to the McDonald criteria. Annals of Neurology 69, 292–302. 10.1002/ana.22366 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Polman CHC, Reingold SCS, Edan GG, Filippi MM, Hartung H-PH, Kappos LL, Lublin FDF, Metz LML, McFarland HFH, O’Connor PWP, Sandberg-Wollheim MM, Thompson AJA, Weinshenker BGB, Wolinsky JSJ, 2005. Diagnostic criteria for multiple sclerosis: 2005 revisions to the “McDonald Criteria”. Annals of Neurology 58, 840–846. 10.1002/ana.20703 [DOI] [PubMed] [Google Scholar]
  23. Rae-Grant A, Day GS, Marrie RA, Rabinstein A, Cree BAC, Gronseth GS, Haboubi M, Halper J, Hosey JP, Jones DE, Lisak R, Pelletier D, Potrebic S, Sitcov C, Sommers R, Stachowiak J, Getchius TSD, Merillat SA, Pringsheim T, 2018. Comprehensive systematic review summary: Disease-modifying therapies for adults with multiple sclerosis. Neurology 90, 789–800. 10.1212/wnl.0000000000005345 [DOI] [PubMed] [Google Scholar]
  24. Reich DS, Lucchinetti CF, Calabresi PA, 2018. Multiple Sclerosis. New England Journal of Medicine 378, 169–180. 10.1056/nejmra1401483 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Savova GK, Masanz JJ, Ogren PV, Zheng J, Sohn S, Kipper-Schuler KC, Chute CG, 2010. Mayo clinical Text Analysis and Knowledge Extraction System (cTAKES): architecture, component evaluation and applications. Journal of the American Medical Informatics Association : JAMIA 17, 507–513. 10.1136/jamia.2009.001560 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Services, C. for M. and M., n.d. ICD-10 General Equivalence Mappings [WWW Document]. undefined. URL https://www.cms.gov/Medicare/Coding/ICD10/index.html (accessed 2.24.21).
  27. Thompson AJ, Banwell BL, Barkhof F, Carroll WM, Coetzee T, Comi G, Correale J, Fazekas F, Filippi M, Freedman MS, Fujihara K, Galetta SL, Hartung H-P, Kappos L, Lublin FD, Marrie RA, Miller AE, Miller DH, Montalban X, Mowry EM, Sørensen PS, Tintoré M, Traboulsee AL, Trojano M, Uitdehaag BMJ, Vukusic S, Waubant E, Weinshenker BG, Reingold SC, Cohen JA, 2018a. Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. Lancet neurology 17, 162–173. 10.1016/s1474-4422(17)30470-2 [DOI] [PubMed] [Google Scholar]
  28. Thompson AJ, Baranzini SE, Geurts J, Hemmer B, Ciccarelli O, 2018b. Multiple sclerosis. Lancet 391, 1622–1636. 10.1016/s0140-6736(18)30481-1 [DOI] [PubMed] [Google Scholar]
  29. Xia Z, Secor E, Chibnik LB, Bove RM, Cheng S, Chitnis T, Cagan A, Gainer VS, Chen PJ, Liao KP, Shaw SY, Ananthakrishnan AN, Szolovits P, Weiner HL, Karlson EW, Murphy SN, Savova GK, Cai T, Churchill SE, Plenge RM, Kohane IS, Jager PLD, 2013. Modeling disease severity in multiple sclerosis using electronic health records. PloS one 8, e78927–e78927. 10.1371/journal.pone.0078927 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Zhang T, Goodman M, Zhu F, Healy B, Carruthers R, Chitnis T, Weiner H, Cai T, Jager PD, Tremlett H, Xia Z, 2020. Phenome-wide examination of comorbidity burden and multiple sclerosis disease severity. Neurology: Neuroimmunology & Neuroinflammation 7, e864. 10.1212/nxi.0000000000000864 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Zhang Y, Cai Tianrun, Yu S, Cho K, Hong C, Sun J, Huang J, Ho Y-L, Ananthakrishnan AN, Xia Z, Shaw SY, Gainer V, Castro V, Link N, Honerlaw J, Huang S, Gagnon D, Karlson EW, Plenge RM, Szolovits P, Savova G, Churchill S, O’Donnell C, Murphy SN, Gaziano JM, Kohane I, Cai Tianxi, Liao KP, 2019. High-throughput phenotyping with electronic medical record data using a common semi-supervised approach (PheCAP). Nature Protocols 14, 3426–3444. 10.1038/s41596-019-0227-6 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material
NIHMS1752830-supplement-Supplementary_Material.docx (3.4MB, docx)

Data Availability Statement

Code for analysis and figure generation is available at <https://github.com/nkim1322/MS_EHR_paper1>. Summary data that support the findings of this study as well as anonymous data used for this study are available upon reasonable request to the corresponding author.

RESOURCES