1 INTRODUCTION

Multiple sclerosis (MS) is a progressive demyelinating disease characterized in the majority by acute attacks, separated by periods of stability.1 The accumulation over time of MS lesions leads to disability and an increased risk of mortality over decades.1, 2 Due to the heterogeneity and the fact that disability often develops long after disease onset, short-term studies often fail to capture the full impact of the disease. Therefore, long-term population-based, observational studies from sources such as registries are required to accurately assess both morbidity and mortality in MS. The validated outcome of mortality reported as, standardized mortality ratio (SMR), remains unambiguous and allows for comparison over time and between populations.

Mortality data may also allow for evaluation of the long-term effects of disease modifying treatments (DMTs) in a real-world setting, the impact of early and effective diagnosis, co-morbidities, lifestyle and environmental factors.3 The SMR, which assesses the death rate, adjusted for age and sex, in the MS population when compared to the general population, assesses mortality risk. However, the SMR is dependent on duration of follow-up and the characteristics of those included in the studies, a factor that is influenced by the method and breadth of data collection.

In 2015, a meta-analysis of the SMR in MS by Manouchehrinia et al. (2015) found a 2.8 times greater risk of mortality in MS patients with no temporal trend over time from 1949–2012.2 Consensus regarding the increased mortality compared to the general population was confirmed in later studies which showed an SMR for patients with MS between 2.45–2.92.3-9 The lack of a temporal trend in the meta-analysis by Manouchehrinia et al. was in contrast to the improved survival over time demonstrated by the study by Koch-Henriksen et al., who found over fifty percent improvement in SMR in a 1990–1999 cohort compared to a 1950–1959 cohort.7 While a study by Lunde et al. also reported a dramatic improvement in SMR of 0.7 in a cohort from 1997–2012.9 This contrasting evidence was later supported by large-center studies in Denmark, Sweden and Norway which included 43,852 patients demonstrating an improved survival over time: citing early diagnosis, disease modifying treatment, improved treatment of co-morbidities and improved rehabilitation as potential causes.7-9 However, differences in mortality trends are subject to a time bias and changing methods of data collection which may explain the dramatic improvements seen in later cohorts.

For this reason, we aimed to update the findings from the meta-analysis by Manouchehrinia et al. (2015), but also to investigate whether there is a temporal trend in all-cause mortality among people with MS. This knowledge can aid in the interpretation of treatment effects on overall mortality of people with MS going forward.

2 MATERIAL AND METHODS

In performing this systematic literature review, we followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) protocol.10

2.1 Search strategy and study selection

A comprehensive literature search was performed using Cochrane Library, Scopus and Medline by two independent investigators. The search strategy for all three databases was with the Mesh terms “Multiple Sclerosis Mortality” and “standardized mortality ratio” or “standardized mortality ratio” for publication in the period 2014–2020. This period was chosen as the previous meta-analysis included all published data up until November 2014. The last date searched was the May 18, 2020. Reference lists were screened for additional papers.

2.2 Eligibility criteria

Original longitudinal cohort studies which reported overall all-cause SMR or age-matched hazard ratio for multiple sclerosis patients when compared to the age- and sex-matched counterparts were included. Studies were required to report total population size, number of deaths and length of study period to be included. Only studies published in English were included. For multiple studies using the same cohort, the study with the longest duration of follow-up that met the inclusion criteria was included.

2.3 Ethics

No ethical approval was needed because data from previously published studies in which informed consent was obtained by primary investigators was retrieved and analyzed.

2.4 Data extraction

Total number of patients, number of deaths, mean follow-up period, person year, study type and duration were extracted from included studies meeting eligibility criteria without additional quality assessments regarding biases or certainty. Each investigator extracted data in a blinded approach from literature databases and references identified in the papers. Each study's corresponding author was contacted to provide information on people years where this information was not published. Where crude mortality ratio (CMR) was not reported, this was calculated using the person years. SMRs with 95% confidence interval (CI) were extracted for all-cause mortality, gender-specific mortality and cause-specific mortality for respiratory, cardiovascular disease, suicide and cancer. If cause-specific mortality was not reported, it was obtained from the relevant publications of the same cohort.

2.5 Data analysis 2.5.1 SMR

Included studies’ SMRs were pooled using inverse variance models for all-cause, cause-specific and sex specific SMR. Log-SMR was used in the analysis as it has a more normal sampling variance that is preferred when the reference populations between studies are different. The fixed effects model was applied.

2.5.2 CMR For studies where CMR was not reported, it was calculated by dividing the number of deaths during a study period by the person year follow-up time. Where person year was not reported it was estimated by mean follow-up time multiplied by number of persons.9 The 95% CI for the CMR was obtained as below. CMRs were pooled by inverse variance model. The random effects model was also applied. 2.5.3 Trends in SMR over time

A meta-regression model was used to determine whether study date was associated with effect size differences with log-SMR being the dependent variable and the middle year of follow-up period being the independent variable.

2.5.4 Heterogeneity

Heterogeneity between studies was evaluated with the I2 statistic. In the case of high heterogeneity, influence analysis was used to establish causes of high heterogeneity.

2.5.5 Publication bias

A funnel plot was used for visual assessment of publication bias. An adjusted rank correlation test was used to complement the funnel graph. Egger's regression test was used for investigation of small study bias.

2.5.6 Sensitivity Analysis

The cause-specific SMR were estimated excluding direct cause of death data to investigate whether cause of death reporting affected the findings.

Statistical analyses were performed using R version 4.0.3 20.20.10 and RStudio version 1.3.1093.

3 RESULTS 3.1 Study selection

A total of 56 studies were identified using the search strategy, with one additional study being identified through study citations. Of these, 18 duplicates were removed. Thirty-nine studies were screened for eligibility criteria through reading of abstracts. Thirty studies were excluded because they either did not include multiple sclerosis, reported on DMT, reported only co-morbidities, reported on incidence or were not longitudinal cohort studies. Nine articles remained for full text analysis of which two were excluded because no overall SMR/HR was reported.11, 12 Seven studies were used in qualitative synthesis and meta-analysis, in addition to the 12 studies identified through our search terms and which were already included in the meta-analysis by Manouchehrinia et al. in 2014 (Figure 1). Table 1 summarizes the 19 included studies. Of these 19 studies, seven were in addition to the previous meta-analysis adding 156,241 new patients and 1,312,210 person years to the original cohort of 27,423 patients and 437,832 person years. A total of 17 studies were included in the CMR meta-analysis. Two studies were excluded as person year could not be established.9, 13

Flow chart of study procedure

TABLE 1. Study characteristics ID Study Patients (n) Deaths (n) Person years Study period Underlying or direct cause of death CMR (95% CI) SMR (95% CI) SMR Female (95% CI) SMR Male (95% CI) 1 Hansen et al. (Denmark)17 9881 4254 207,862 1949–1997 Underlying 20.46 (19.85–21.7) 2.89 (2.81–2.98) 3.14 (3.01–3.17) 2.66 (2.54–2.78) 2 Sadovnick et al. (Canada)13 2348 115 1972–1985 NA 2.0 (1.63–2.36) 3 Torklidsen et al. (Norway Hordeland County)15 878 198 35,120 1953–2005 Direct 5.63 (4.85–6.42) 2.66 (2.31–9.8) 3.11 (2.58–3.27) 2.23 (1.81–9.12) 4 N. Koch-Henriksen et al. (Denmark)7a 12,847 6102 339,483 1950–2015 NA 17.98 (17.53–18.42) 2.45 (2.39–2.51) 2.53 (2.41–2.65) 2.36 (2.25–2.48) 5 Lunde et al. (Norway Hordeland County)9a 1388 291 1953–2012 Direct 2.7 (2.4–3.0) 2.9 (2.903–3.4) 2.5 (2.1–2.9) 6 Ragonese et al. (Italy)19b 183 30 4188 1960–2007 NA 7.16 (4.6–9.71) 2.12 (1.32–3.46) 2.22 (1.23–4.0) 2 (0.89–4.46) 7 Smestad et al. (Norway- Oslo)18 386 263 12,172 1972–2005 Direct 6.8 (5.9–7.7) 2.47 (2.09–6.38) 3.4 (3–3.9) 2.2 (1.9–6.16) 8 Sumelahti et al. (Finland)27 1614 219 32,165 1971–2006 NA 21.6 (19.02–27.22) 2.8 (2.6–6.15) 2.94 (2.36–3.62) 2.02 (1.56–6.17) 9 Leray et al. (France)19 1879 68 23,906 1976–2003 NA 2.84 (2.15–3.52) 1.3 (1–8.18) 1.5 (1.1–2.1) 1.1 (0.8–8.9) 10 Burkill et al. (Sweden)8 a,b, a,b 29,617 9563 358,366 1968–2012 Underlying 26.68 (25.78–27.59) 2.92 (2.86–2.99) 2.06 (2.97–3.15) 2.75 (2.65–2.84) 11 Kingwell et al. (Canada)28 6917 1025 77,950 1980–2007 Underlying 13.14 (12.34–13.94) 2.88 (2.71–3.07) 3.01 (2.79–3.25) 2.68 (2.43–2.96) 12 Hirst et al. (Wales)29 366 218 5609 1985–2006 Underlying 38.86 (33.8–43.92) 2.79 (2.44–3.18) 1.65 (1.55–1.76) 1.3 (1.21–1.4) 13 Leary et al. (France)3a 27,603 1569 420,801 1980–2012 NA 3.79 (3.55–3.92) 1.48 (1.41–1.55) 3.14 (2.67–3.69) 2.26 (1.79–2.85) 14 Rodríguez-Antigüedad Zarranz et al. (Spain)20 1283 89 16,422 1987–2011 NA 5.41 (4.29–6.54) 2.78 (2.2–3.38) 2.73 (1.94–3.51) 3.26 (2.27–4.24) 15 Kingwell et al. (Canada)6a 6629 1416 104,236 1986–2013 Underlying 13.58 (12.88–14.29) 2.71 (2.55–2.87) 2.79 (2.61–2.98) 2.53 (2.32–2.76) 16 Sandi et al. (Hungary)4a 740 121 10,303 1993–2013 Underlying 11.74 (9.66–13.82) 2.52 (2.1–3.01) 2.57 (2.03–3.2) 2.46 (1.82–3.25) 17 Manouchehrinia et al. (UK)30 923 80 18,717 1994–2012 Not specified 4.27 (3.34–5.2) 1.99 (1.7–2.33) 1.8 (1.4–2.3) 2.41 (1.95–2.69) 18 Lalmohamed et al. (UK)15 1270 69 3937 2001–2008 Direct 17.52 (13.42–21.62) 3.51 (2.63–4.69) 3.94 (2.73–5.68) 2.96 (1.84–4.77) 19 Foloun et al. (France)5a,b, a,b 78,805 1080 78,805 2012–2013 NA 13.7 (12.88–14.58) 2.56 (2.41–2.72) 2.55 (2.5–2.75) 2.58 (2.34–2.83) Note: Person years estimated from mean follow up period. Abbreviations: CMR, crude mortality ratio; SMR, standardised mortality ratio. a Studies in addition to previous meta-analysis. b Calculated 95% CI. Used in cause specific SMR analysis. Cause of death definitions were not applicable in those studies that did not investigate this.

A pooled total of 185,557 individuals with 26,770 deaths and 1,750,042 person year follow-up were included in these analyses. Person year follow-ups ranged from 3.9/1000 in United Kingdom (UK) to 420/1000 person years in France. All observational cohorts were informed with register data and from high-income settings.

3.2 CMR

The pooled CMR was 13.45/1000 person years (95% CI 9.56 to 17.34, I2 = 100%, p = 0) (Figure 2).

Forest plot of crude mortality ratio sored by cohort's follow-up period midpoint year (*studies added to updated meta-analysis)

3.3 SMR

The SMR ranged from 1.30 (95% CI 0.45 to 3.72) in France to 3.51 (2.63; 4.69) in the UK.14, 15 The pooled SMR from the fixed effects model for all-cause mortality was 2.61 (95% CI 2.58 to 2.65, I2 = 98%, p < .01) (Figure 3). When stratified by sex, the pooled SMR was 2.47 for men (95% CI 2.42 to 2.52, I2 = 95%, p < .01) and 2.57 (95% CI 2.53 to 2.61, I2 = 98%, p < .01) for women.

Forest plot of the pooled standardized mortality ratio (SMR) data, arranged by the respective cohort's midpoint year in the follow-up period (*studies added to updated meta-analysis)

3.4 Heterogeneity in SMR

Measures of heterogeneity confirmed a substantial proportion of variability between studies (I2 = 98%, Q = 760, t2 = 0.038, p < .01). The prediction interval for SMR ranged from 1.63 to 3.83. Causes of high heterogeneity were determined by influence analysis (Figure S1). Four studies3, 7, 8, 16 were excluded due to a higher overall SMR of 2.66 (95% CI 2.58 to 2.75, I2 = 63%) with moderate heterogeneity.

3.5 Cause-specific SMR

The pooled cause-specific SMRs are presented in Figure 4. Eleven studies in total reported cause-specific SMR: six of the studies.1, 10-12, 15, 16 used the underlying cause of death as their definition for cause-specific SMR, four used direct cause of death3, 5, 7, 17 and one did not specify.18

Forest plot of causes specific standardized mortality rates (SMR) due to cardiovascular disease, respiratory illness and infection, accident and suicide, and cancer. SMR, standardized mortality ratio; CVD, cardiovascular disease; *studies in addition to previous meta-analysis. ^accident and suicide. (A) accident only

3.5.1 Cardiovascular disease

A total of 10 studies reported SMR for cardiovascular disease. Of these 10 studies, four were in addition to the previous publication.4, 6, 8, 9 The SMR for cardiovascular disease was 1.74 (95% CI 1.67 to 1.81, I2 = 94%, p < .01).

3.5.2 Respiratory illness and infection

Eight studies reported an SMR for respiratory illness and infection. Of these eight studies, three were new to the cohort of the previous meta-analysis.6, 8, 9 The SMR for respiratory illness and infection was 4.70 (95% C.I. 4.54 to 4.87, I2 = 97%, p < .01) in the main analyses and 4.60 (95% CI 4.43 to 4.8) when direct cause data was removed in a sensitivity analysis.

3.5.3 Accident and suicide

Eleven Studies reported an SMR for accident and/or suicide. Two studies reported on accident only,6, 8 whereas four studies reported on both accident and suicide.9, 15, 17, 18 The remaining five studies commented on suicide only. The pooled SMR for accident and suicide was 1.81 (95% CI 1.64 to 2.00, I2 = 86%, p < .01). In a subset analysis where studies that reported on accident only were removed, the SMR was 2.13 (95% C.I. 1.86 to 2.44, I2 = 85%, p < .01).

3.5.4 Cancer

A total of 10 studies reported on cancer. Of the 10 studies, three were in addition to the previous meta-analysis.4, 6, 9 The SMR for cancer was 0.99 (95% CI 0.93 to 1.06, I2 = 94%, p < .01).

3.6 Time trends

Using meta-regression models, we were not able to reject the null hypothesis of no effect of time on SMR. This was true for both all-cause SMR (co-efficient 0.001 (change in log-SMR/year), 95% CI −0.0099 to 0.0102, p = .98) (Figure 5) and CMR (co-efficient −0.0655 (deaths/1000 person/year, 95% CI −0.43 to 0.3, p = .72). For males, the co-efficient was 0.0012 (CI −0.011 to 0.013, p = .85) and for females, −0.0036 (95% CI −0.15 to 0.01, p = .53).

Bubble plot of meta-regression model with log standardized mortality ratio (SMR [log]) as the independent variable and midpoint of follow-up as the dependent variable. Numbers are used as study identifiers (Table 1) 3.7 Publication bias

Visual inspection of the funnel plot did not reveal asymmetry in the result of all-cause SMR (Figure S1). Further inspection with rank correlation test (p = .25) and Egger's test (p = .33) supported this.

4 DISCUSSION

In this study, we carried out a meta-analysis on the SMR for MS, updating results from a previous 2015 meta-analysis. A SMR is a valuable measure to identify the apparent risk of dying at a given age, given that the patient has also been diagnosed with MS. The SMR is calculated using the total number of observed deaths divided by the total number of expected deaths for patient's age- and sex-matched counterparts. If a SMR is greater than one, this suggests that there are more deaths in the studied group than would be expected for the general population. If the SMR is consistent over time, it suggests that any secular change in population survival over time is shared by both the MS group and general population. A total of seven large-center studies have been published since the meta-analysis by Manouchehrinia et al. (2015) was completed. Inclusion of the recent studies increased the total patient numbers to 160,000 compared to 27,423 in the previous meta-analysis. In this study which included the significant increase person years, we found the overall SMR for MS patients was 2.6, which suggests MS patients have an almost three times greater risk of death compared to the general population. This rate is similar to what was found by Manouchehrinia et al, (2015) with an SMR of 2.8 and suggests the SMR for MS has been stable over time.

One limitation identified in the previous meta-analysis was that 79% of the cohort was made up of Danish and Canadian cohorts, our meta-analysis expanded this cohort with seven additional studies to include 58% from France and a further 27% from Sweden and Denmark, improving the generalizability of results for high-income countries.2 SMR results between the two meta-analysis have varied between 2.8 and 2.6. The slightly higher SMR in the earlier study may be attributable to the heavy weighting of the Denmark study by Hansen et al., an issue that was overcome by the addition of later Denmark study by Koch-Henriksen that included a the more recent period of time, 1997–2015.7 The CMR results were not significantly different between the two studies.

In regard to the SMR trends over time, we found no support of a reduced all-cause SMR for people with MS compared to counterparts over the last 65 years. Accordingly, mortality has decreased among people with MS at similar rates to the general population likely due to advances in modern medicine and lifestyle improvements. Although this is consistent with the systematic review by Manouchrinia et al, (2015), it contrasts with what single-population-based studies by Lunde et al (2017), Burkill et al (2017) and Koch-Henriksen (2017).2,