Effectiveness of first generation disease-modifying therapy to prevent conversion to secondary progressive multiple sclerosis

Background: The use of disease-modifying therapies (DMTs) in multiple sclerosis (MS) has been associated with reduced relapse rates and accumulation of disability. However, studies examining impact of DMT on risk of transition to secondary progressive MS (SPMS) leveraging population-based nationwide data are still rare. Here, we determine the population incidence of conversion to SPMS using two consecutive nation-wide cohorts, one immediately before and one after the introduction of DMT in Sweden. Methods: We included two consecutive population cohorts of relapsing-remitting MS (RRMS) from the Swedish national MS register for the periods 1975 – 1994 ( n = 2161), before DMT availability, and 1995 – 2011 ( n = 3510), in which DMTs, mainly first generation DMT (injectables), became available and eventually were used by 70% of patients. We explored the risk of transition to SPMS as a calendar year function encompassing the two cohorts. In addition, we determined the incidence of transition to SPMS through age strata below and above 50 years in untreated and treated patient subgroups. Results: The risk of conversion to SPMS (adjusted for current age, current time since onset, calendar year and sex) was significantly lower in the second compared with the first population cohort (hazard ratio 0.58; CI 0.48, 0.70). The risk of SPMS conversion per calendar year decreased by 2.6% annually ( p < 0.001) after 1995. The risk of SPMS conversion increased with age until age 50. Thereafter, it was unchanged or decreased among those with early MS onset age ( < 35 years), but continued to increase with onset at higher age, with similar trends in treated and untreated subgroups. Conclusion: The incidence of SPMS conversion significantly decreased at the population level after introduction of first generation DMTs by 1995. DMT efficiency was confirmed by a downward turn of the annual trajectory of the risk of SPMS conversion after 1995. An onset age determined pattern of variable SPMS incidence in higher age appeared in both treated and untreated


Introduction
Several centres have observed a milder long-term course of multiple sclerosis (MS) over the last decades (Kister et al., 2012), (Beiki et al., 2019).Although the cause of this change is complex, including improved diagnostic techniques, modified diagnostic criteria, and changed demographics, the disease-modifying therapies (DMTs), introduced in Sweden 1995 (Sorensen et al., 2020), are considered to have impact on MS course.However, the evidence based on different criteria of long-term outcomes associated with first generation DMT (injectables) is inconsistent (Shirani et al., 2012).Conversion to SPMS is a key determinant of long-term disease evolution and prognosis (Scalfari et al., 2013).Most observational follow-up studies with at least 5 years of follow-up showed delay of conversion to SPMS after treatment with first generation DMT (Bergamaschi et al., 2016;Trojano et al., 2007;Tedeholm et al., 2007;Drulovic et al., 2013;Signori et al., 2016;Zhang et al., 2015).Investigators were increasingly aware of several inherent biases in observational studies, the most detrimental of which probably is indication bias, which means that patients with more severe disease move into the treated subgroup, confounding indicators of effectiveness (Sormani and Bruzzi, 2015) , (Kalincik and Butzkueven, 2016).Methods used to manage indication bias in observational studies include regression analyses with validated predictors (Tedeholm et al., 2013), or propensity score analysis (Trojano et al., 2007) , (He et al., 2020).If predictors are not available, an alternative design may be based on two successive population cohorts, one right before the introduction of DMT became available and one immediately following with DMT in general use (Veugelers et al., 2009).The SPMS incidence in the defined population is a sum of treated and untreated patients' SPMS incidences, so the total population-based incidence is expected to evade indication bias, although it may be affected by other bias.We here compare the risk of transition to SPMS in two consecutive population cohorts, including the age segment after 50 years which was generally excluded from trials (Vollmer et al., 2021), one immediately before DMT became available in Sweden, and one right after the introduction of first generation DMT.

Study design
We herein (A) compare the incidence of transition to SPMS in two consecutive population materials, an untreated cohort from 1975 to 1994 and a partially DMT treated cohort from 1995 to 2011.The first cohort included only DMT untreated patients, and the second included all DMT treated and untreated patients with onset in this and previous cohorts.(B) Covering the same periods, we study the calendar year incidence of transition to SPMS annually from 1950 to 2011.Furthermore, we (C) study the incidence of transition to SPMS in DMT treated and untreated subgroups defined by onset 1950-1994 or 1995-2011 (incidence cohorts) and with individual age below or above 50 and (D) compare subgroups only defined by DMT treatment and adjusted for current age, time since onset and current calendar year.

Patient materials
Before the DMT era, the therapeutic tradition in Sweden was conservative.Except for sporadic use of azathioprine and monthly pulsed infusions of methylprednisolone, MS patients received no long-term immunosuppressive treatment prior to 1995, the year we terminated the DMT untreated cohort.For our partially treated cohort we were faced with a choice between a long follow-up vs a homogenous first generation DMT treated material; as a compromise we terminated the cohort at December 31st 2011, when second-generation DMTs started to come into wider use.Although natalizumab (Miller et al., 2003) was approved in 2006, its use was initially limited (10% of time on therapy until 2012; Supplementary Table 1).For 1950-2011, Swedish National MS contained 17,971 patients (Fig. 1) who had provided consent for inclusion (www.neuroreg.se).The data extraction 2017 encompassed 12,246 patients with either RRMS or SPMS and with disease onset before 2012, of whom 10,492 had at least one visit with information on treatment status.To achieve uniform data quality, we restricted the cohort to patients from the 12 largest Swedish MS centres, with data from more than half of the patient population of the Swedish national register.Ultimately, we imported data from 6500 patients with onset during 1950-2011 from the Swedish MS register on August 30, 2017.We used the Swedish national MS register's SPMS criterion of an insidious increase in neurological deficit, typically initiated or dominated by a pyramidal syndrome, which is compatible with the Lublin and Reingold 1996 consensus definition of SPMS (Lublin and Reingold, 1996).The active centres checked data for quality, reducing the rate of missing data on the year of transition to SPMS to 1.3%.There was consensus among participating centres that the uncertainty of retrospective estimates of the year of onset of incipient progression in the national register amounts to a few years.We included individual data on sex, date of birth, age, date of MS onset, date of each visit, date when treatment was started, type of DMT, all DMT periods, interval from disease onset to start of first DMT, and year of transition from RRMS to SPMS.From the imported data we defined the following cohorts: 1975-1994 resp. 1995-2011 The study was approved by the Research Ethics Committee of Gothenburg (Dnr 545-16, August 30, 2016) and by the Research Board of the Swedish National MS Register (Dnr 44).

Statistical methods
To estimate the instantaneous hazard function (HF) for the transition from RRMS to SPMS, we used a modified Poisson regression model (Albertsson-Wikland et al., 2016;Breslow and Day, 1987;Skoog et al., 2014).We assessed the relationship between the risk of SPMS and sex, current time since disease onset, current age, and current calendar year.The observation period for each patient was divided into 1-month intervals.Outcome was transition from RRMS to SPMS per individual.The time at risk was censored at the year of conversion to SPMS or date of last visit.The Poisson variable is continuously updated and provides a momentary risk, allowing us to include current age in addition to time since onset, calendar year and risk of SPMS in the model.The meaning of momentary risk is more evident from our previous study using Poisson regression in a web-based predictor estimating the current risk of SPMS at any time during RRMS (Skoog et al., 2019).

Comparing population cohorts (substudy A)
Some patients with onset during 1975-1994 started DMT later, during 1995-2011.When comparing the SPMS risk in the MS population living during 1995-2011 vs living during 1975-94 independent of their onset time, the modified Poisson regression model described above was used, resulting in a HR with 95% confidence interval.The HR were also adjusted for current age, current time since onset, current calendar year and sex (Table 2a and b).

Risk of transition to SPMS in relation to calendar year (substudy B)
All patients, irrespective of onset year, age and treatment, H. Tedeholm et al. contributed to a HF describing the incidence of transition from RRMS to SPMS at the current calendar year.The HF of SPMS was linear in the two variables time since onset and current age.Current calendar years were treated as piece-wise continuous linear variables with a breakpoint at the calendar years of 1980 and 1995.This HF was used to calculate the HR describing the increase or decline with one year of change in current calendar year with 95% CI.The risk of SPMS was also investigated with spline functions with current calendar year as continuous variable using breakpoints in current calendar year.Breakpoints for current calendar year were 1975, 1990, and 2005.HF was used to calculate the incidence of SPMS with 95% CIs with sex, time since onset, current age, and current calendar year as covariates.

Risk of SPMS relative to age and age at onset (substudy C)
Separate models were developed for individuals with onset during 1950-1994 and 1995-2011 and by intervals of age at onset, and by therapy (Tables 3-4).When applying the model to untreated individuals, the observation time was censored at first medication intake (first or second generation DMT).When including a treated individual, the observation time started at the first medication.Thus, one individual could contribute to the model for both untreated and treated individuals with observation time before and after treatment.The HF of SPMS was linear in the two variables time since onset and current age.Current age was treated as piece-wise continuous linear variables with a breakpoint at the age of 50 years (Table 4).Among patients with onset during 1995-2011, those with an age of onset of <35 years generally did not reach 50 years of age because of the short follow-up to the end of 2011.Consequently, 30 and 40 years of age were substituted for 50 years as the breakpoints in these groups.Hazard ratios (HRs) to describe increase or decrease with age were calculated from HF. Two-sided p values were used for all analyses, with p < 0.05 considered significant.
The risk of SPMS was also investigated with spline functions in current age as continuous variables using breakpoints (Figs.2-3).The splines were second order functions between the breakpoints and linear H. Tedeholm et al. functions at the tails, resulting in a smooth curve.Available breakpoints for Figs. 2 and 3 were age 25,35,45,and 55. Breakpoints (knots) define different regions (or partitions) for age.For some subgroups the data do not cover all breakpoints.Then the breakpoints for current age were determined based on the range of data (Supplementary Table 2).HF was used to calculate the incidence of SPMS with 95% confidence intervals (CIs) for different analyses (Figs.2-3, significance values shown), with sex, time since onset, current age, and current calendar year as covariates.

Risk of transition to secondary progression in relation to treatment (substudy D)
To study the overall effect of treatment on the risk of SPMS we used the variable "treatment" as a time-dependent covariate, adjusted for current age, time since onset and current calendar year.

Test of risk of transition from RRMS to SPMS between the first and second population cohort (substudy A)
HR for the difference in risk of transition to SPMS between population Cohort 1 (all participants living through the calendar years 1975-1994) and the population Cohort 2 (all participants living through the calendar years 1995-2011) was 1.18 unadjusted, and 0.58 adjusted for current age, current time since onset, current calendar year and sex (Table 2).The large difference between unadjusted and adjusted values mainly depends on adjustment for current age and calendar year.

Risk of transition to SPMS relative to calendar year (substudy B)
The risk of transition to SPMS was calculated as a function of calendar year 1950-2011 and adjusted for current time since onset, current age, and sex.The risk of SPMS in untreated patients showed a tendency to increase by 3.0% per year during 1980-1995 (HR 1.030, 95% CI: 1.016-1.043,p < 0.001).On the contrary, the risk diminished by 2.6% annually after 1995 (HR 0.975, 95% CI: 0.966-0.984,p < 0.001; Fig. 4).

Lifetime risk of SPMS (trend before and after 50 years of age) (substudy C)
3.3.1.Risk of SPMS before and after age 50 in the incidence cohort with onset 1950-1994, censored December 31, 1994, untreated , untreated In this subgroup the risk of conversion to SPMS increased significantly with current age by 6%-9% annually among those < 50 years of age (p < 0.004).The risk decreased significantly after 50 years of age in the subgroup with youngest age at onset (< 25 years, p = 0.02), with no significant change with age in the subgroups with older age at onset (≥ 25 years; Table 4, representative example with confounders locked at the indicated values in Fig. 2a).

Risk of SPMS before and after age 50 in the incidence cohort with onset 1950-1994, censored at treatment or 2011, untreated but with DMT available from 1995
The risk of conversion to SPMS increased also in this subgroup significantly with current age by 5%-9% annually before the age of 50 (p < 0.001).The risk for SPMS after 50 years of age decreased 4% per year for those with MS onset at age < 25 years (p = 0.074) or 25-35 years (p = 0.051; Table 4, representative example with confounders locked at the indicated values in Fig. 2b).4, representative example with confounders locked at the indicated values in Fig. 3a).

Risk of SPMS before and after age 50 (or substituted by age 30 or 40 due to insufficient follow-up time to age 50 with early onset) in an incidence cohort with onset 1995-2011, DMT treated
In this subgroup (where a majority initiated DMT), we observed a yearly increased risk of transition to SPMS with current age before 50 years of age (p < 0.05).After 50 years of age, we observed a continuous increase in the risk of SPMS in those with age of onset ≥35 years (p < 0.0069) and no significant change after the breakpoint with younger age at onset (Table 4, representative example with confounders locked at the indicated values in Fig. 3b).DMT treated (1995DMT treated ( -2011) )

vs. untreated patients (substudy D)
Treatment with DMTs, when analysed as a time-dependent covariate, adjusted for current age, time since onset and current calendar year, was associated with an increased risk of transition to SPMS in both Cohort 1 (after 1995) (HR 1.37, 95% CI: 1.20-1.56,p < 0.001) and Cohort 2 (HR 1.45, 95% CI: 1.16-1.80,p < 0.001).The HR decreased with longer latency to start of treatment, and the higher incidence in treated periods was consistent in each of the age at onset groups (<25, 25-35, >35 years).

Discussion
We here demonstrate that the incidence of transition from RRMS to SPMS was significantly lower in a 1995-2011 population cohort, right after the introduction of disease-modifying therapy (DMT) in Sweden 1995, than in the immediately preceding 1975-1994 untreated cohort.
The 1995-2011 RRMS patients had a treatment option, mainly with first generation DMT (injectables), and were gradually treated up to more than 70%.The annual incidence of SPMS demonstrated a sharp reduction around 1995, the year when the first DMT was approved, after adjusting for factors influencing the calendar year incidence.A similar incidence of SPMS in treated and untreated subgroups before and after age 50 may be due to indication bias, which probably also explains the paradoxical result of DMT treatment by direct comparison of treated vs untreated patients, despite adjusting for time since onset.Recommendations followed by most neurologists in Sweden from 1995 restricted the use of DMT to active disease, probably reinforcing indication bias.In order to evade indication bias (Sormani and Bruzzi, 2015) , (Kalincik and Butzkueven, 2016), we used a calendar year-based design of comparing two consecutive populations, the first untreated and the next including both untreated patients and patients starting DMT.With this population-based design, the overall SPMS incidence should be less influenced by preferential movement of severe cases to the treated subgroup.However, the comparisons between two consecutive populations cohorts may be influenced by other biases, notably competing risks.Exogenous risk factors for MS possibly influencing the SPMS incidence such as low vitamin D levels or smoking (Manouchehrinia    (Gout, 2008).
The population-based design using a common parameter from an untreated and a subsequent treated population was previously used in a Canadian disability progression study demonstrating a moderate longterm effect of introduction of first generation DMT (Veugelers et al., 2009).A similarly designed study used two cohorts with contemporary treated vs. historical untreated controls, and concluded that biases producing exaggerated outcomes with historical controls would equate biases producing underestimated outcomes with contemporary controls  (Shirani et al., 2012).We suggest that indication bias and competing risks had opposing effects in the present study.A population-based study with similar aim as the present demonstrated a sharp increase in the age at disability milestones in a cohort diagnosed after 2000, 5 years after the introduction of DMTs, and a further, sharper positive change in this parameter attributed to the introduction of second-generation DMTs (Capra et al., 2017).
In the 1950-1995 cohort we observed an increasing risk of conversion to SPMS during the first decades of their disease, followed by decades with the risk leveling out or decreasing, which probably indicates the presence of a neurodegenerative response with induction in a window of early MS-related inflammatory focal lesions, analogous to reported MRI and neuropathology data in the early vs. later stages of MS (Frischer et al., 2009).Age-related degeneration may contribute to the increased risk of transition to SPMS with higher age at onset.The occurrence of a middle-age maximum in cases with a younger age of onset argues against the idea of MS as a primary degenerative disorder, in which a relentless tendency for conversion to the SPMS phenotype with increasing age would be expected.Although the present age-dependent outcomes before and after age 50 were similar in the treated and untreated strata, possibly due to indication bias, the middle-age maximum of risk of conversion to SPMS suggests that anti-inflammatory therapy may reduce the risk of SPMS if instituted in the early phases of MS.However, the residual risk of conversion to SPMS does not abate after age 50, indicating that trials on DMT discontinuation in this age group could use SPMS, in addition to markers of current activity, as a meaningful outcome.
The present demonstrates, in agreement with a majority of variously designed long-term follow up studies (Goodin et al., 2012), that first generation DMT have a long-term moderate efficacy, such as the DMT-induced 4 to 7 years delay of SPMS conversion demonstrated in registry studies from Italy and Sweden (Trojano et al., 2007) , (Tedeholm et al., 2013).First generation DMT (injectables) are still used by approximately 7% of MS patients in Sweden, with a trend towards abandoning these DMT, particularly for escalation therapy in younger patients (VAP, www.neuroreg.se).While matched studies showed superior efficacy of specific second generation DMT over first generation (injectables) (Brown et al., 2019), further parallel studies at the population level using the present calendar year based design might explore an associated overall reduction of SPMS incidence and a possible delay of its age maximum.

Conclusion
Altogether, the efficacy of first generation DMT (injectables) to decrease the SPMS incidence is robust at the population level, and detectable despite complex confounders at the individual level, however their efficacy is moderate.

Fig. 1 .
Fig. 1.Flowchart of patient selection for the present study.The material is derived from the Swedish MS register and includes patients with RRMS onset during 1950-2011.
et al., 2014)  are not expected to elicit a significant change in the SPMS risk trajectory during a few years.More relevant, the McDonald diagnostic criteria first published 2001(McDonald et al., 2001) may have reduced the subsequent apparent SPMS risk.However, we contend that the introduction of DMT from 1995 was a major contributor to the downturn of the SPMS incidence trajectory after 1995.A further type of probable confounding in the present study is survivorship bias, here apparently lowering the incidence in the first part of the 1950-95 cohort, as severe cases with onset from 1950 into the 1980s were likely lost when patients died or moved to nursing homes.Unfortunately, the present material extending over decades lacks data on predictors essential for useful regression or propensity score methods

Fig. 2 .
Fig. 2. 2a.Hazard function for SPMS in patients with onset during 1950-1994 and censoring in Dec 1994 in three age at onset groups.The incidence of SPMS is a function of current age and adjusted for time from onset, calendar year, and gender.The figure shows a representative example of a man for whom time after onset is fixed to 5 years and the calendar year to 1990.2b.Group 1b Hazard function for SPMS in untreated patients with onset during 1950-1994 and censoring at first treatment.The incidence of SPMS is a function of current age and adjusted for time from onset, calendar year, and gender.The figure shows a representative example of a man for whom time after onset is fixed to 5 years and the calendar year to 1990.

Fig. 3 .
Fig. 3. 3a.Hazard function for SPMS in patients with onset during 1995-2011 and censoring at first treatment in three age at onset groups.The incidence of SPMS is a function of current age and adjusted for time from onset, calendar year, and gender.The figure shows a representative example of a man for whom time after onset is fixed to 5 years and the calendar year to 2000.3b.Hazard function for SPMS in patients with onset during 1995-2011 and censoring at 2011, with only treated periods included.The incidence of SPMS is a function of current age adjusted for time from onset, calendar year, and gender.The figure shows a representative example of a man for whom time after onset is fixed to 5 years and the calendar year to 2000.

H Tedeholm :
Conceptualization, Formal analysis, Writingoriginal draft.F Piehl: Data curation, Writingreview & editing.J Lycke: Resources, Data curation, Writingreview & editing.J Link: Formal analysis.L Stawiarz: Formal analysis.J Burman: Data curation.P de Flon: Data curation.K Fink: Data curation.M Gunnarsson: Data curation.J Mellergård: Data curation.P Nilsson: Data curation.P Sundström: Data curation, Writingreview & editing.A Svenningsson: Data curation, Writingreview & editing.H Johansson: Formal analysis, Writingoriginal draft.O Andersen: Conceptualization, Project administration, Resources, Writingoriginal draft.Declaration of Competing Interest Helen Tedeholm received a research grant from the Swedish MS Society in 2018.Funds for research time were granted by the Western Swedish Region.Fredrik Piehl has received research grants from Genzyme, Merck KGaA, and UCB, and fees for serving as chair of DMC in clinical trials with Parexel.Jan Lycke has received travel support and/or lecture honoraria from Biogen, Novartis, Merck, Alexion, Roche, and SanofiGenzyme; has served on scientific advisory boards for Biogen, Novartis, Merck, Roche, BSM, and SanofiGenzyme; serves on the editorial board of Acta Neurologica Scandinavica; and has received unconditional research grants from Biogen and Novartis.Jenny Link has nothing to declare.Leszek Stawiarz has nothing to declare.Joachim Burman has nothing to declare.Pierre de Flon has nothing to declare.Katarina Fink has received honoraria for lectures and advisory boards from Merck, Novartis, Roche, Biogen, and Aktelion.Martin Gunnarsson has nothing to declare.Johan Mellergård has received honoraria for advisory boards from Biogen, Sanofi Genzyme and Merck and lecture honoraria from Merck.Petra Nilsson has received travel support from Bayer Schering Pharma, Merck Serono, Biogen, and Genzyme a Sanofi Company; honoraria for lectures and advisory boards from Merck Serono and Genzyme a Sanofi Company; serves on advisory boards for Novartis and Roche; lectures for Biogen; and has received unrestricted grants from Biogen.Peter Sundström has nothing to declare.Anders Svenningsson has nothing to declare.Helena Johansson has nothing to declare.Oluf Andersen serves on the editorial board of Acta Neurologica Scandinavica.

Table 1
Registry-based patient data.

Table 2 a
. SPMS incidence in Cohort 1 and 2. In this subgroup, conversion to SPMS risk increased with current age before the age breakpoints of 30-50 years (p < 0.044).Thereafter, the yearly risk for SPMS increased after age 50 with current age for those with onset ≥ 35 years of age, with no significant change after the breakpoint for those who were <35 years of age at onset (p < 0.001; Table

Table 3
SPMS incidence in three cohorts, unadjusted.

Table 4
Risk factors for hazard function for SPMS incidence *. * Rationale for substituted breakpoints at 30 and 40 years of age, see text. *