Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2018 May 1.
Published in final edited form as: Environ Res. 2018 Feb 6;163:43–52. doi: 10.1016/j.envres.2018.01.040

Ozone, NO2 and PM10 are associated with the occurrence of multiple sclerosis relapses. Evidence from seasonal multi-pollutant analyses

Maxime Jeanjean a, Marie-Abele Bind c, Jonathan Roux a,b,d, Jean-Claude Ongagna e, Jérôme de Sèze e, Denis Bard a, Emmanuelle Leray a,b,d,*
PMCID: PMC5886008  NIHMSID: NIHMS949398  PMID: 29426027

Abstract

Background

Triggers of multiple sclerosis (MS) relapses are essentially unknown. PM10 exposure has recently been associated with an increased risk of relapses.

Objectives

We further explore the short-term associations between PM10, NO2, benzene (C6H6), O3, and CO exposures, and the odds of MS relapses’ occurrence.

Methods

Using a case-crossover design, we studied 424 MS patients living in the Strasbourg area, France between 2000 and 2009 (1783 relapses in total). Control days were chosen to be±35 days relative to the case (relapse) day. Exposure was modeled through ADMS-Urban software at the census block scale. We consider single-pollutant and multi-pollutant conditional logistic regression models coupled with a distributed-lag linear structure, stratified by season (“hot” vs. “cold”), and adjusted for meteorological parameters, pollen count, influenza-like epidemics, and holidays.

Results

The single-pollutant analyses indicated: 1) significant associations between MS relapse incidence and exposures to NO2, PM10, and O3, and 2) seasonality in these associations. For instance, an interquartile range increase in NO2 (lags 0–3) and PM10 exposure were associated with MS relapse incidence (OR = 1.08; 95%CI: [1.03–1.14] and OR = 1.06; 95%CI: [1.01–1.11], respectively) during the “cold” season (i.e., October-March). We also observed an association with O3 and MS relapse incidence during “hot” season (OR = 1.16; 95%CI: [1.07–1.25]). C6H6 and CO were not significantly related to MS relapse incidence. However, using multi-pollutant models, only O3 remained significantly associated with the odds of relapse triggering during “hot” season.

Conclusion

We observed significant single-pollution associations between the occurrence of MS relapses and exposures to NO2, O3 and PM10, only O3 remained significantly associated with occurrence of MS relapses in the multi-pollutant model.

Keywords: Multiple sclerosis, Relapse, Air pollution, Socioeconomic position

1. Introduction

Multiple Sclerosis (MS) is the most frequent neuro-inflammatory disease of the central nervous system (CNS), affecting almost 2.3 million people worldwide (Browne et al., 2014). The prevalence in France is about 1.5 individuals for 1000 (Foulon et al., 2017). In about 85% of cases, patients experience relapse(s) (Browne et al., 2014), that is, patients experience exacerbations of neurologic disabilities followed by partial or complete remission. Relapses characterize the relapsing-remitting MS course.

The etiology of MS remains unclear, yet clearly multifactorial (McKay et al., 2016). Genetic predispositions (HLA-DRB1, IL2RA and IL7R most associated genes) may represent only one quarter of estimated heritability of MS (Sawcer et al., 2014). Main environmental factors found to be influencing susceptibility to experience an MS onset are Epstein-Barr virus infection, tobacco consumption (either passive or active) and reduced levels of vitamin D increase or low sunlight exposure. In addition, women appear to be at greater risk than males (Ascherio et al., 2014; Belbasis et al., 2015; Lauer, 2010). Relapses predisposing risk factors have been investigated to a lesser extent and associations reported are discordant, namely young age, short MS duration, low serum vitamin D levels, smoking, psychological and other stress, vaccination, infections, post-partum, breast feeding and assisted reproduction (McKay et al., 2016; Kalincik, 2015).

Relapses’ incidence varies across seasons (Damasceno et al., 2012; Fonseca et al., 2009; Handel et al., 2011; Harding et al., 2017; Jin et al., 2000; Ogawa et al., 2004; Salvi et al., 2010; Spelman et al., 2014; Tremlett et al., 2008), suggesting a possible role of season-dependent factors such as meteorological variables (Watad et al., 2016) and ambient air pollution (Angelici et al., 2016; Oikonen et al., 2003; Roux et al., 2017).

A few studies have investigated the role of air pollution in the development of MS, reporting mixed results (Chen et al., 2017; Gregory et al., 2008; Heydarpour et al., 2014). In Teheran, significant clustered patterns (p<.001) and difference in exposure to sulfur dioxide (SO2), PM10, NO2 and nitrogen oxides (NOx) were observed in MS cases compared to random controls (Heydarpour et al., 2014). Gregory et al. (2008) also suggested that PM10 might play a role in MS onset (Gregory et al., 2008). However, these studies suffered from a number of limitations (e.g. small sample size, imprecise exposure assessment etc.). Recently, Chen et al. (2017) observed, using a Cox proportional hazards model adjusted for individual features and latitude, that proximity living to heavy traffic was not associated with a higher incidence of MS (Chen et al., 2017). Concerning relapses, some studies investigated the link between air pollution and the odds of relapses triggering (Tremlett et al., 2008; Oikonen et al., 2003; Roux et al., 2017; Mehrpour et al., 2013; Vojinović et al., 2015), MS-related hospitalization (Angelici et al., 2016) and MS inflammatory activity (Bergamaschi et al., 2017). Associations were observed in several studies between PM10 exposure and MS relapse risk (Angelici et al., 2016; Oikonen et al., 2003; Roux et al., 2017; Bergamaschi et al., 2017). Among studies that examined possible influence of other air pollutants such as CO, O3, SO2 and NOx (Tremlett et al., 2008; Oikonen et al., 2003; Mehrpour et al., 2013), one study reported exposure to acidic gas (NO2, NO and SO2) to be associated with risk of relapse (Oikonen et al., 2003). Overall, the authors did not take into consideration pertinent confounding factors such as meteorological parameters, infections etc.

There is a need to consider multi-pollutant models in environmental epidemiology, especially in studies relating MS occurrence to air pollutants’ concentrations. The environmental health-related associations are complex to investigate in observational studies due to the high correlations and possible interactions between pollutants, as well as the seasonality of the pollutants’ concentrations. To our knowledge, no study has yet considered multi-pollutant analyses to estimate the associations between air pollution and MS.

In the present study, we estimate the associations of PM10 and other ambient air pollutants (NO2, benzene (C6H6), CO and O3) using a multi-pollutant models to address the confounding issue due to the high correlations between pollutants originating from the same source.

2. Materials and methods

2.1. Study design

We conducted a time-stratified case-crossover design to explore the associations between short-term air pollutants variations (i.e., PM10, NO2, C6H6, O3, and CO) and multiple sclerosis relapses occurrences (Maclure, 1991). This study design consists in within-subject comparisons by selecting for each patient his/her own control, i.e., the air pollutant exposure levels of the same patient will be compared between days of relapse onset (case) and days without any relapse (control). A time-stratified approach was chosen to define control days (Janes et al., 2005). That is, control days were chosen to be±35 days relative to the case (relapse) day. This 35-days interval choice was motivated by the relapse clinical definition which confined a minimum of 30 days between two relapses. Every control day between the 30th and 35th day was excluded, justifying the reason why 158 case days only have a single control instead of two.

Time-invariant or long-term varying confounders, such as individual characteristics and behaviors, are controlled through within-subject comparisons. This approach also permits to tackle time trends such as seasonality, between days of the week variation, and the temporal autocorrelation (Janes et al., 2005).

2.2. Study setting

The Strasbourg Metropolitan Area (SMA), located in North-Eastern France, aggregates 28 municipalities, 21 of which are rural and seven urban. The SMA sprawls over 316 km2 with 450,000 inhabitants. It is subdivided into 186 French census blocks, defined as a sub-municipal division and designed by French National Institute for Statistics and Economic Studies (INSEE). Census blocks are devised according to land use, homogeneity of population size and socio-economic features. They are the smallest spatial unit in France for which socio-economic data are made available due to French confidentiality regulations. In average, a census block is 2000 inhabitants (ranging from 2 to 4885), with a surface from 0.05 km2 up to 19.6 km2.

2.3. Study population and environmental data

2.3.1. Patients’ and relapses’ inclusion criteria

Patients’ data were provided by the multiple sclerosis network alSacEP based in Alsace North-Eastern region, since 2006. All patients were managed through the European Database for Multiple Sclerosis (EDMUS) using a standardized definition and management of patients’ data (Confavreux et al., 1992).

Study period was January 1, 2000 to December 31, 2009. Inclusion criteria for patients were: i) clinical definition fitting McDonalds’ MS criteria; ii) first symptoms of MS occurred before December 31th, 2009; iii) patients were affected with relapsing-remitting and secondary progressive forms; iv) residence address within Strasbourg Metropolitan Area.

When the day of relapse occurrence was doubtful (uncertain or unknown), the relapse (i.e., case day) was excluded, leading to additional patients exclusions.

The French Authority for Data Confidentiality (CNIL) approved the present protocol (DR-2015-504).

2.3.2. Air pollution data

Air pollution concentrations of PM10, NO2, C6H6, O3, and CO were estimated throughout the study period on an hourly basis at the census block scale. The deterministic ADMS-Urban air dispersion model was used considering different parameters, namely background pollution measurements, emissions inventories, meteorological data but also land use or surface roughness (Atmospheric Dispersion Modeling System) (Carruthers et al., 2000). Details of parameters and model performance have been previously discussed (Havard et al., 2009; Bard et al., 2014). Previous works have shown that air pollution assessment performance proved excellent results (Laurent et al., 2008): coefficients for the modeled and effectively measured ambient concentrations were highly correlated 0.87 for NO2, 0.73 for PM10 and 0.84 for O3.

We did not assess the plausible role of sulfur dioxide (SO2) which was suggested in the recent literature (Heydarpour et al., 2014) because i) Strasbourg Metropolitan Area concentrations are low (≤ 11 μg/m3) in the Alsace region, and ii) it originates from a single location, which altogether decrease the preciseness of the modeling. Concentrations of PM2.5, which represents a substantial proportion of PM10 (Marcazzan et al., 2001), were not measured routinely during the study period. PM10 is a proxy measure of PM2.5. The potential interventions to reduce PM10 and PM2.5 involve similar sources. However, benzene, rarely measured by air pollution monitoring systems, was properly measured and modeled in our study setting.

2.3.3. Established or likely confounding variables

According to the literature on the link between air pollution and different health outcomes occurrence and especially MS, we considered different time-varying confounders in our study. Meteorological parameters (daily temperature, relative humidity and atmospheric pressure) were obtained from the French meteorological service (Météo France). Daily pollen counts were provided by the National Network of Aerobiological Surveillance (Thibaudon, 2004). Weekly influenza-like case count was given by the “Sentinelles” network (Flahault et al., 2006) of the French National Institute of Health and Medical Research (INSERM). We considered holidays, which could potentially influence industrial activities and road traffic, as well as stress level, fatigue or being at home or not.

2.4. Data analysis

Associations between exposures to air pollutants and the occurrence of relapses were estimated by fitting distributed-lag linear models within conditional logistic regressions. We examined how the associations between lagged exposures and the outcomes varied across lags. This methodology, previously developed for the analysis of time-series data (Gasparrini et al., 2010), was performed here in the context of case-crossover data.

Seasonal variations occur for both air pollutants concentrations and MS incidence (Harding et al., 2017). Therefore, we fitted regression models separately for “hot” (April 1st to September 30th) and “cold” seasons (October 1st to March 31th). We estimated odds ratios corresponding to an interquartile range (IQR) increment of concentration (μg/m3). An association was considered significant if the p-value was less or equal to 0.05. All statistical analyses were performed using R software (v. 3.2.3) (R Core Team, 2014) and the “dlnm” R package (Gasparrini, 2011).

An unconstrained (i.e. nonparametric) lag structure was used for the initial analysis. Then, we considered another choice of distributed-lag function that assumes a constant lag effect within days. The constant lag modeling prevents overfitting of the data. This assumption is equivalent to fitting a constrained model that includes consecutive daily exposure moving average. Because some pollutants were highly correlated, we fitted multi-pollutant models to determine which pollutant( s) explained the single-pollutant associations. The collinearity between the independent variables of the multi-pollutant models was assessed with the Generalized Variance Inflation Factor (GVIF(1/[2df])) (Fox and Monette, 1992). To assess collinearity, values of GVIF were compared to the threshold of 10, which was considered as a maximum value according to the literature (O’brien, 2007). Each model was adjusted on all lagged (including lag 0) daily concentrations and daily temperature to also take into account the exposure correlation between the lags.

3. Results

3.1. Individual demographic and clinical patients’ characteristics

We obtained carefully verified data for 1783 relapses and selected 3408 control days from 424 patients. Data were analyzed separately for “cold” and “hot” seasons (i.e., 888 case days and 1703 control days during the “hot” season and 895 case days and 1705 controls during the “cold” season) over the 2000–2009 study period. All of them were living in 145 French census blocks of the Strasbourg Metropolitan Area (SMA) at some time over the study period.

Patients’ characteristics are presented in Table 1. Sex ratio (Female: Male) was 2.93 and patient’s mean age at MS clinical onset was 30.5 (±10.0) years old. Throughout the study period, patients experienced in average 4.2 relapses (2.11 in cold season and 2.09 in hot season).

Table 1.

Baseline characteristics of the 424 patients included in the study.

Multiple Sclerosis patients N = 424
Demographic characteristics
Gender
Males 108 (25.5%)
Females 316 (74.5%)
Clinical characteristics
Mean age at MS onset (±sd), years (30.5±10.0)
Mean follow-up duration (±sd), years (6.6±3.5)
Mean relapses per patient (between 2000 and 2009) (4.2±4.7)
MS form (at last information)
Relapsing-Remitting 352 (83.0%)
Secondary Progressive 72 (17.0%)
Open in a new tab

sd: standard deviation.

Most of the patients included were affected with relapsing-remitting MS form (83.0%). The others were affected with secondary progressive form (17.0%).

3.2. Environmental data and flu-like infections

A description of environmental data is detailed in Table 2. Over the study period, mean concentrations of PM10, NO2, C6H6, and CO in the SMA were higher during “cold” than “hot” season. By contrast, mean concentrations of O3 were higher during “hot” than “cold” seasons (respectively 86.8±30.9 μg/m3 and 37.3±20.2 μg/m3, p<001). As expected, pollutants were highly correlated (Table 3, Fig. 1). Concentrations of PM10, NO2, CO and benzene were highest in the center of the CUS and concentrations of O3 were highest in the periphery (Fig. 1). We observed seasonal variations in pollen counts (higher in the “hot” season) and flu-like infections (higher in the “cold” season) (Table 2).

Table 2.

Baseline characteristics of daily air pollutants concentrations, meteorological features, pollen and infections across Strasbourg (France) Metropolitan Area, 145 census blocks (2000–2009).

Parametersb “Hot” seasona
“Cold” seasona
Mean Sdc Min. Q1 Median Q3 Max. Mean Sd Min. Q1 Median Q3 Max.
Air pollutants (μg/m3)
PM10 18.94 6.78 4.68 13.88 17.97 22.98 51.59 23.70 12.69 2.55 14.52 21.59 30.12 101.50
NO2 29.30 11.66 4.06 20.95 28.16 36.45 92.31 38.66 13.90 3.32 28.88 38.60 48.02 95.90
C6H6 1.10 0.59 0.14 0.71 0.97 1.34 4.67 2.23 1.24 0.19 1.33 2.02 2.91 10.58
CO 577.7 58.54 501.9 536.8 565.1 603.7 998.3 608.1 84.26 502.7 546.4 585.9 647.8 1175.0
O3 86.85 30.89 15.12 65.49 83.44 105.20 221.00 37.26 20.19 2.44 20.47 35.66 51.45 107.20
Meteorological features
Maximum temperature (°C) 22.61 5.56 5.10 19.10 22.80 26.50 38.50 9.45 6.02 −10.30 5.38 9.60 13.52 28.50
Maximum atmospheric pressure (hPa) 1010.01 10.41 979.10 1001.00 1011.00 1019.00 1031.00 1013.00 13.03 970.30 1003.00 1013.00 1023.00 1043.00
Maximum relative humidity (%) 91.99 5.49 59.0 90.00 93.00 96.00 100.00 93.86 4.35 72.00 92.00 95.00 97.00 100.00
Allergy
Pollen count (daily mean number of grains/m3) 130.30 210.10 0.00 21.16 69.84 158.10 2368.00 40.04 130.55 0.00 0.00 0.00 15.64 1519.00
Infections
Influenza-like epidemics (Nb of cases per week) 48.73 107.87 0.00 0.00 0.00 59.00 851.00 453.50 587.02 0.00 112.00 254.00 535.00 3486.00
Open in a new tab

Wilcox.test (hot vs. cold): Significant (p<.001) for all the PM10, NO2, C6H6, O3 and CO pollutants.

a

“Hot” season (April to September) and “Cold” season (October to March).

b

Estimations for the 145 census blocks of Strasbourg Metropolitan Area.

c

Sd: Standard deviation.

Table 3.

Air pollutants daily mean concentrations value correlation coefficient (r), Strasbourg (France) Metropolitan Area, 145 census blocks (2000–2009).

PM10
NO2
C6H6
CO
O3
“cold” “hot” “cold” “hot” “cold” “hot” “cold” “hot” “cold” “hot”
PM10 “cold” 1
“hot” 1
NO2 “cold” 0.71 1
“hot” 0.70 1
C6H6 “cold” 0.72 0.64 1
“hot” 0.50 0.59 1
CO “cold” 0.55 0.72 0.59 1
“hot” 0.46 0.59 0.57 1
O3 “cold” −0.39 −0.48 −0.54 −0.48 1
“hot” 0.29 −0.06 −0.18 −0.21 1
Open in a new tab

Correlation coefficient (r): Pearson’r.

Fig. 1.

Fig. 1

Open in a new tab

Ambient air pollution concentrations and SES index across Strasbourg Metropolitan Area census blocks (2000–2009).

3.3. Relation between air pollutants and the occurrence of relapses

3.3.1. Unconstrained distributed-lag single pollutant models

Fig. 2 presents the associations between air pollutants concentrations (for every lag of the week preceding the relapse) and the occurrence of relapses, separately for “hot” and “cold” seasons. We observed significant negative and positive associations on several distinct days, when adjusting individually on lagged daily air pollutants concentrations, lagged daily maximum temperature, day of relapse maximum relative humidity, maximum atmospheric pressure, as well as pollen count, influenza-like epidemics, and holidays.

Fig. 2.

Fig. 2

Open in a new tab

Associations between exposure to air pollutants concentrations (lags 0–6 days before relapse onset, inconstant lag effect) and risk of MS relapse triggering, according to season. a”Hot” season (April to September) and “Cold” season (October to March); Odds-ratio (lags 0–6 days, inconstant lag effect). OR corresponds to an interquartile increment of concentration and are represented with their 95% Confidence Interval. Multivariate conditional logistic regression models were adjusted on all lagged daily air pollutants concentrations, daily maximum temperature, day of relapse maximum relative humidity, maximum atmospheric pressure, pollen count, influenza-like epidemics, and holidays.

We observed a significantly increased risk (about 40%) with NO2 exposure at lag 1 in “hot” season, but no association during “cold” season. For PM10, we found an increased risk at lag 1 in “hot” season (OR = 1.26 [1.03–1.54]) whereas lag 2 was significant in “cold” season (OR = 1.28 [1.05–1.55]). Exposure to benzene was significantly associated with 30% excess risk in the “cold” season only, at lag 2 (OR = 1.32 [1.05–1.67]). We found that CO exposure increased the odds of MS relapse to a lesser extent in both “cold” (20%, lag 3) and “hot” season (30%, lag 2). The strongest association (roughly 60%) was observed with O3 exposure during “hot” season at lag 2. However, we noticed significant inverse associations at lag 3 in “hot” season for NO2 (OR = 0.82 [0.68–0.99]) and for benzene (OR = 0.81 [0.67–0.98]), which motivated the multi-pollutant approach presented in Section 3.3.3.

3.3.2. Constrained distributed-lag single pollutant models

We also assumed a constant lag effect within days (Table 4 and Fig. 3). Finally, we checked whether the associations we observed were still significant when considering a longer period (i.e. lags 0–6, one-week period) (Table 4). As compared to the unconstrained models, we observed a much lower increased risk (less than 10%) for both NO2 and PM10 and only in “cold” season for lag 0–3 (respectively, OR = 1.08 [1.03–1.14] and OR = 1.06 [1.01–1.11]) (Table 4 and Fig. 3). In contrast, O3 exposure was significantly associated with an excess MS relapse risk of 16% in “hot” season only, lag 0–3 (OR = 1.16 [1.07–1.25]). Exposure to C6H6 yielded borderline significant excess risk estimates, during “cold” season, lag 0–3 (OR = 1.05 [1.00–1.10]) and inversely in “hot” season (OR = 0.98 [0.94–1.02]). We found no association with CO exposure. Associations with PM10 and NO2 exposures were no longer significant when one week of exposure (lag 0–6) was considered (respectively, OR = 1.02 [0.99–1.05]; and OR = 1.03 [1.00–1.07]) (Table 4). However, associations with O3 remained significant although weaker when one week of exposure was considered (OR = 1.13 [1.07–1.19]).

Table 4.

Estimated risk for associations between exposure to air pollutants concentrations (lags 0–3 days) and risk of MS relapse triggering, according to season.

Parameter (μg/m3)d “Hot” season (n=2594)a
“Cold” season (n = 2597)a
ORb CI95%c OR CI95%
PM10 Lags 0–3 1.04 [0.99–1.09] 1.06 [1.01–1.11]
Lags 0–6 (one week) 1.03 [1.00–1.06] 1.02 [0.99–1.05]
NO2 Lags 0–3 1.01 [0.96–1.07] 1.08 [1.03–1.14]
Lags 0–6 (one week) 1.02 [0.98–1.05] 1.03 [1.00–1.07]
C6H6 Lags 0–3 0.98 [0.94–1.02] 1.05 [1.00–1.10]
Lags 0–6 (one week) 0.99 [0.96–1.02] 1.02 [0.99–1.05]
CO Lags 0–3 0.96 [0.90–1.02] 1.04 [0.98–1.10]
Lags 0–6 (one week) 0.97 [0.93–1.01] 1.02 [0.99–1.06]
O3 Lags 0–3 1.16 [1.07–1.25] 0.96 [0.90–1.03]
Lags 0–6 (one week) 1.13 [1.07–1.19] 0.98 [0.94–1.02]
Open in a new tab
a

“Hot” season (April to September) and “Cold” season (October to March) / n = Number of cases and control days.

b

Odds-ratio (lags 0–3 days and lags 0–6 days, constant lag effect). OR corresponds to an interquartile increment of concentration.

c

95% Confidence Interval. Multivariate conditional logistic regression models were adjusted on all lagged daily air pollutants concentrations, daily maximum temperature, day of relapse maximum relative humidity, maximum atmospheric pressure, pollen count, influenza-like epidemics, and holidays.

Fig. 3.

Fig. 3

Open in a new tab

Associations between exposure to air pollutants concentrations (lags 0–3 days before relapse onset, constant lag effect) and risk of MS relapse triggering, according to season.

3.3.3. Constrained distributed-lag multiple-pollutant models

Table 5 displays the multi-pollutant analyses results using the same constrained adjusted models as above. We estimated the associations between the odds of relapses using the all-pollutant model (MFull/p=5) and PM10 exposure during “cold” season. The estimate was no longer significant (OR = 1.02 [0.95–1.11]) as compared to the results from single pollutant models (Table 4). The NO2-MS relapse associations were borderline significant (OR = 1.08 [1.00–1.18]). In the “hot season”, only the PM10-MS relapse association remained borderline significant (OR = 1.07 [1.00–1.15]).

Table 5.

Multi-pollutant analysis.

PM10
NO2
C6H6
CO
O3
“Hot”a “Cold”a “Hot” “Cold” “Hot” “Cold” “Hot” “Cold” “Hot” “Cold”
ORb CI95%c ORb CI95%c OR CI95% OR CI95% OR CI95% OR CI95% OR CI95% OR CI95% OR CI95% OR CI95%
Multi-pollutant (p = 4)d
MnoPM10 / p=4 – – 1.07 [0.98–1.17] 1.10 [1.02–1.18] 0.97 [0.92–1.02] 1.02 [0.96–1.08] 0.94 [0.84–1.04] 0.97 [0.89–1.05] 1.13 [1.04–1.23] 1.01 [0.93–1.10]
M noNO2 / p=4 1.08 [1.03–1.13] 1.06 [0.99–1.14] – – 0.96 [0.93–0.99] 1.01 [0.94–1.08] 0.95 [0.90–1.00] 1.00 [0.92–1.08] 1.12 [1.01–1.24] 0.99 [0.92–1.07]
MnoC6H6 / p=4 1.05 [0.98–1.13] 1.03 [0.96–1.10] 1.01 [0.91–1.11] 1.09 [1.00–1.18] – – 0.93 [0.84–1.04] 1.01 [0.93–1.09] 1.12 [1.04–1.22] 0.96 [0.89–1.04]
M noCO / p=4 1.07 [1.00–1.15] 1.02 [0.94–1.10] 0.98 [0.91–1.06] 1.07 [0.99–1.16] 0.95 [0.90–1.00] 1.00 [0.94–1.08] – – 1.14 [1.05–1.23] 1.02 [0.94–1.11]
MnoO3 / p=4 1.09 [1.01–1.17] 1.02 [0.95–1.11] 1.04 [0.95–1.15] 1.08 [1.00–1.18] 0.95 [0.90–1.00] 1.00 [0.94–1.08] 0.89 [0.81–0.99] 0.96 [0.89–1.04] – –
All-pollutant (p=5)e
MFull / p=5 1.07 [1.00–1.15] 1.02 [0.95–1.11] 1.02 [0.92–1.13] 1.08 [1.00–1.18] 0.95 [0.90–1.01] 1.00 [0.94–1.08] 0.94 [0.85–1.05] 0.96 [0.89–1.04] 1.12 [1.03–1.22] 1.01 [0.93–1.10]
Open in a new tab
a

“Hot” season (April to September) and “Cold” season (October to March) / n = Number of cases and control days.

b

Odds-ratio concentrations (lags 0–3 days, constant lag effect). OR corresponds to an interquartile increment of concentration.

c

95% Confidence Interval. Multivariate conditional logistic regression models were adjusted on all lagged daily air pollutants concentrations, daily maximum temperature, day of relapse maximum relative humidity, maximum atmospheric pressure, pollen count, influenza-like epidemics, and holidays.

d

Multi-pollutant models (p = 4: four pollutants included in the model).

e

All-pollutant models (p = 5: all five pollutants included in the model).

When the model was fitted with all the pollutants except NO2, the risk of relapses with PM10 exposure became significant in “hot” season (OR = 1.08 [1.03–1.13]) as compared to the single pollutant analysis (Table 4) where no association was observed. An equivalent magnitude of risk was observed from single to multi-pollutant models with O3 exposure. When the model was run excluding PM10, the risk of relapse associated with NO2 exposure in “cold” season was very close to the single pollutant results. While benzene and CO exposures in “hot” season were not associated with risk in the single pollutant analyses, we observed inverse significant associations using multi-pollutant models: for benzene, excluding NO2, OR = 0.96 [0.93–0.99]; for CO, excluding O3, OR = 0.89 [0.81–0.99].

Collinearity between pollutant variables across multi-pollutant models was considered as moderate (all GVIF(1/[2df]) < 10, and most of them<2) (Table 6).

Table 6.

Generalized Variance Inflation Factor values for pollutants by multi-pollutant models.

GVIF(1/[2df])a
“Hot”b “Cold”b
Multi-pollutant (p=4)c
MnoPM10 / p=4 NO2 1.93 1.74
C6H6 1.39 1.58
CO 1.80 1.66
O3 2.26 2.06
M noNO2 / p=4 PM10 1.68 2.07
C6H6 1.44 1.87
CO 1.43 1.58
O3 2.25 2.05
MnoC6H6 / p=4 PM10 1.91 1.92
NO2 2.26 1.91
CO 1.79 1.68
O3 2.29 2.05
M noCO / p=4 PM10 1.98 2.24
NO2 1.81 1.79
C6H6 1.45 1.87
O3 2.14 1.98
MnoO3 / p=4 PM10 1.96 2.27
NO2 2.25 1.90
C6H6 1.46 1.85
CO 1.68 1.61
All-pollutant (p=5)d
MFull / p=5 PM10 1.98 2.27
NO2 2.28 1.92
C6H6 1.45 1.87
CO 1.80 1.68
O3 2.29 2.06
Open in a new tab
a

GVIF(1/[2df]): Generalized Variance Inflation Factor; Coefficient used to assess collinearity.

b

“Hot” season (April to September) and “Cold” season (October to March).

c

Multi-pollutant models (p=4: four pollutants included in the model).

d

All-pollutant models (p=5: all five pollutants included in the model).

4. Discussion

We have shown significant associations between short-term exposure to ambient air pollutants (NO2, PM10 and O3) and the occurrence of MS relapses in a French population-based study spanning 10 years (2000–2009) (Table 4). We confirmed our previous results (Roux et al., 2017) on PM10-associated risk using a more sound statistical approach. In particular, our model-adjustments were more precise, i.e., we consider all lagged (including day 0–3 instead of 1–3) of daily concentrations and daily temperature instead of the single day of relapse, and therefore took into account the correlation between the lag exposures. We estimated odds ratios corresponding to an interquartile range (IQR) increment of concentration (μg/m3) instead of a one-unit increment of ln PM10. These differences in methodology might explain the variation of magnitude observed between both studies (present analysis OR = 1.06 [1.02–1.11]: previous OR = 1.40 [1.08–1.81]) (Roux et al., 2017). We investigated the relation between air pollutants and the occurrence of relapses on each day of a one-week period before onset, which is supposed to be the maximum air pollution effect (Angelici et al., 2016). When fitting the models considering one week of exposure to air pollutants, we observed no association beyond lag 3 (except a “protective” effect for CO at lag 5 in “hot” season) (Fig. 2). These observations yield to conduct analysis considering only 3 days (instead of 6) before the occurrence of relapses and we did not observe any major difference (results not shown). Our results are in line with those of the literature (Angelici et al., 2016; Oikonen et al., 2003; Bergamaschi et al., 2017) which reported a short term association between PM10 and the odds of relapse (Angelici et al., 2016; Bergamaschi et al., 2017).

Since air pollutants are highly correlated and that correlations vary across season (Table 3) due to complex reactions such as photochemical reactions which necessitate sunlight, we explored the impact of air pollution testing multi-pollutant models (Table 5). Several associations decreased in magnitude when adding pollutants in the model, It might be due to the relation between pollutants, such as collinearity (Yoo et al., 2014). Only O3 was systematically significantly associated with MS risk during “hot” season, using one, four, or five pollutants in the model. However, variations observed by single pollutant to multi-pollutant models were very limited in size. O3 showed the largest association with the odds of MS relapse. In this study, we provide pollutantspecific comparisons. Results of PM10 in the multi-pollutant models were partly confounded by NO2 and vice versa. Indeed, the both pollutants are highly correlated for both seasons (around 0.70). We did not conduct any bi-pollutant analyses because we considered that it did not reflect actual exposures and that adding statistical tests might lead to observe false positives which could alter the interpretation. Collinearity may be an important issue in multi-pollutant models, so that we have calculated the GVIF^(1/(2df)) for each pollutant included in the multipollutant distributed-lag models. As coefficient values were low (< 5), all multi-pollutant associations were reported.

Air pollutants’ concentrations exhibit seasonal patterns, leading us to conduct analyses according to season (“cold”: from October to March; “hot”: from April to September) rather than fitting models adjusting for season as in the Italian study (Angelici et al., 2016). Farez et al. (2015) observed that melatonin, which is a neuro-hormone regulated by seasonal variation in sunlight level and especially night length, was inversely correlated with relapses incidence (Farez et al., 2015). Yet, some studies have reported associations between UV level and MS exacerbations (Spelman et al., 2014; Tremlett et al., 2008), suggesting a possible role of erythemal ultraviolet radiation in the production of serum vitamin D production that might influence relapse incidence (Tremlett et al., 2008). However, Harding et al. (2017) reported highest risks in early summer when sunshine duration is elevated and lowest risks at the end of summer (Harding et al., 2017), suggesting that vitamin D might not be the only environment factor incriminated in MS relapses (e.g., interactions etc.). Our results seem to corroborate seasonal changes in MS activity, as associations with air pollutants varied across season, so as established or likely confounding variables we accounted in our models (meteorological, infections and allergy variables). Indeed, we observed differences in risks between air pollutants concentrations and risk of relapses according to “hot” and “cold” seasons, yet this categorization is only an uncertain proxy of UV exposure. Moreover, PM10 and NO2 concentrations were associated with MS relapses occurrence only in “cold” months, while O3 was associated in “hot” months. Seasonal changes in air pollutants concentrations, the type of area (e.g., topography, level of traffic etc.) and the seasonal possible individual air pollutant effect on MS activity could explain mixed results concerning seasonal pattern in relapses rate observed in the literature (Harding et al., 2017). We adjusted on holidays since they may influence industrial activities and road traffic but also and particularly the presence of patients at home. We have previously shown that holidays suggested a “protective” association during “cold” months (Roux et al., 2017). This result could be related with a drop in stress-related work (Artemiadis et al., 2011) or a leave for areas featuring different exposure patterns. In addition, during summer, patients are more likely to stay outside and might be more exposed to ozone.

4.1. Limitations

We considered patients’ living address only, which can be a limitation as much they could spend a part of their time out of home (e.g., commuting, job site, leisure…).

However, patients affected with MS generally see their mobility reduced along with the course of the disease. Therefore, there are expected to spend more time at home with time (Kobelt et al., 2009). We did not adjust for sunlight exposure, which is reported to be related to MS incidence (Harding et al., 2017).

4.2. Strengths

This study has several strengths. First, we conducted multi-pollutant models to determine pollutant-specific associations despite the high collinearity between pollutants. Second, a state-of the-art case ascertainment through the systematic reporting of patients followed up within the EDMUS database (Confavreux et al., 1992). Because the data collection started before the study period, most of the relapses were prospectively collected and dates of relapses occurrence were set by neurological experts. We also used a robust and accurate exposure assessment at the census blocks scale, limiting exposure misclassification to the extent possible. Finally, we adjusted models with a number of confounders known to be season-dependent.

The role of air pollution in the pathogenesis of MS remains to be fully elucidated. Recently, Esmaeil Mousavi et al. (2017) formulated the assumption that air pollution might impact MS incidence and activity through biological mechanisms inducing neuroinflammatory-oxidative cascades reactions, decrease of immunological self-tolerance and neurodegeneration (i.e. axonal deterioration and neuronal loss) that finally conduct to autoimmunity (Esmaeil Mousavi et al., 2017). For instance, some of those mechanisms enrolled could be the blood brain barrier breakdown, a mitochondrial dysfunction, an overproduction of free radicals or the expression of inflammatory factors. Moreover, showing a significant association between PM10 concentrations before magnetic resonance imaging (MRI) examination and MRI Gadolinium-enhancing lesions, an Italian study provided recent additional evidence that ambient air pollution might be a determinant for MS inflammatory relapses triggering (Bergamaschi et al., 2017). Epigenetic changes in autoimmune disease also occur, especially changes in DNA methylation (Javierre et al., 2011).

5. Conclusion

Using a precise exposure spatio-temporal model and a clinically-diagnosed outcome from a fairly exhaustive multiple sclerosis registry, we reported associations between exposures to air pollutants (NO2, PM10 and O3) and the risk of MS relapses, when pollutants were assessed individually. In a multi-pollutant model, only O3 was significantly associated with the risk of MS triggering. When assessing the link between exposure to PM10 and MS relapses, we recommend adjusting for NO2 level and vice versa. Taken together, these findings enhanced our understanding of the plausible association between air pollution exposure and MS relapses but further research is needed to confirm this hypothesis. Our observation of an association with PM10 is in line with the results of the few studies published so far. Yet, the association we observed with NO2 and O3 is to our knowledge unprecedented.

Acknowledgments

We are grateful to Dorian Bankovic, Estelle Le Pabic, Thomas Senger, Carole Berthe, and all the neurologists contributing to the alSacEP network.

Funding sources

This work was supported by the EHESP School of Public Health and the alSacEP network. Research reported in this publication was also supported by the Office of the Director, National Institutes of Health under Award Number DP5OD021412. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Written consent from subjects participating in the study was received prior to the study. Moreover, the study was approved by the ethics committee in accordance with the French law.

Footnotes

Conflicts of interest

The authors declare no conflicts of interest.

References

  1. Angelici L, Piola M, Cavalleri T, Randi G, Cortini F, Bergamaschi R, et al. Effects of particulate matter exposure on multiple sclerosis hospital admission in Lombardy region, Italy. Environ Res. 2016;145:68–73. doi: 10.1016/j.envres.2015.11.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Artemiadis AK, Anagnostouli MC, Alexopoulos EC. Stress as a risk factor for multiple sclerosis onset or relapse: a systematic review. Neuroepidemiology. 2011;36(2):109–120. doi: 10.1159/000323953. [DOI] [PubMed] [Google Scholar]
  3. Ascherio A, Munger KL, White R, Köchert K, Simon KC, Polman CH, et al. Vitamin D as an early predictor of multiple sclerosis activity and progression. JAMA Neurol. 2014;71(3):306–314. doi: 10.1001/jamaneurol.2013.5993. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bard D, Kihal W, Schillinger C, Fermanian C, Ségala C, Glorion S, et al. Traffic-related air pollution and the onset of myocardial infarction: disclosing benzene as a trigger? A small-area case-crossover study. PLOS One. 2014;9(6):e100307. doi: 10.1371/journal.pone.0100307. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Belbasis L, Bellou V, Evangelou E, Ioannidis JPA, Tzoulaki I. Environmental risk factors and multiple sclerosis: an umbrella review of systematic reviews and meta-analyses. Lancet Neurol. 2015;14(3):263–273. doi: 10.1016/S1474-4422(14)70267-4. [DOI] [PubMed] [Google Scholar]
  6. Bergamaschi R, Cortese A, Pichiecchio A, Berzolari FG, Borrelli P, Mallucci G, et al. Air pollution is associated to the multiple sclerosis inflammatory activity as measured by brain MRI. Mult Scler Houndmills Basingstoke Engl. 2017;1:1352458517726866. doi: 10.1177/1352458517726866. [DOI] [PubMed] [Google Scholar]
  7. Browne P, Chandraratna D, Angood C, Tremlett H, Baker C, Taylor BV, et al. Atlas of multiple sclerosis 2013: a growing global problem with widespread inequity. Neurology. 2014;83(11):1022–1024. doi: 10.1212/WNL.0000000000000768. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Carruthers DJ, Edmunds HA, Lester AE, McHugh CA, Singles RJ. Use and validation of ADMS-Urban in contrasting urban and industrial locations. Int J Environ Pollut. 2000;14(1/2/3/4/5/6):364. [Google Scholar]
  9. Chen H, Kwong JC, Copes R, Tu K, Villeneuve PJ, van Donkelaar A, et al. Living near major roads and the incidence of dementia, Parkinson’s disease, and multiple sclerosis: a population-based cohort study. Lancet Lond Engl. 2017;389(10070):718–726. doi: 10.1016/S0140-6736(16)32399-6. [DOI] [PubMed] [Google Scholar]
  10. Confavreux C, Compston DA, Hommes OR, McDonald WI, Thompson AJ. EDMUS, a European database for multiple sclerosis. J Neurol Neurosurg Psychiatry. 1992;55(8):671–676. doi: 10.1136/jnnp.55.8.671. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Damasceno A, Von Glehn F, de Deus-Silva L, Damasceno BP. Monthly variation of multiple sclerosis activity in the southern hemisphere: analysis from 996 relapses in Brazil. Eur J Neurol. 2012;19(4):660–662. doi: 10.1111/j.1468-1331.2011.03543.x. [DOI] [PubMed] [Google Scholar]
  12. Esmaeil Mousavi S, Heydarpour P, Reis J, Amiri M, Sahraian MA. Multiple sclerosis and air pollution exposure: mechanisms toward brain autoimmunity. Med Hypotheses. 2017;100:23–30. doi: 10.1016/j.mehy.2017.01.003. [DOI] [PubMed] [Google Scholar]
  13. Farez MF, Mascanfroni ID, Méndez-Huergo SP, Yeste A, Murugaiyan G, Garo LP, et al. Melatonin contributes to the Seasonality of multiple sclerosis relapses. Cell. 2015;162(6):1338–1352. doi: 10.1016/j.cell.2015.08.025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Flahault A, Blanchon T, Dorléans Y, Toubiana L, Vibert JF, Valleron AJ. Virtual surveillance of communicable diseases: a 20-year experience in France. Stat Methods Med Res. 2006;15(5):413–421. doi: 10.1177/0962280206071639. [DOI] [PubMed] [Google Scholar]
  15. Fonseca AC, Costa J, Cordeiro C, Geraldes R, de Sá J. Influence of climatic factors in the incidence of multiple sclerosis relapses in a Portuguese population. Eur J Neurol. 2009;16(4):537–539. doi: 10.1111/j.1468-1331.2008.02528.x. [DOI] [PubMed] [Google Scholar]
  16. Foulon S, Maura G, Dalichampt M, Alla F, Debouverie M, Moreau T, et al. Prevalence and mortality of patients with multiple sclerosis in France in 2012: a study based on French health insurance data. J Neurol. 2017 Mai; doi: 10.1007/s00415-017-8513-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Fox J, Monette G. Generalized collinearity diagnostics. J Am Stat Assoc. 1992;87(417):178–183. [Google Scholar]
  18. Gasparrini A. Distributed lag linear and non-linear models in R: the package dlnm. J Stat Softw. 2011:1–20. [PMC free article] [PubMed] [Google Scholar]
  19. Gasparrini A, Armstrong B, Kenward MG. Distributed lag non-linear models. Stat Med. 2010;29(21):2224–2234. doi: 10.1002/sim.3940. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Gregory AC, Shendell DG, Okosun IS, Gieseker KE. Multiple sclerosis disease distribution and potential impact of environmental air pollutants in Georgia. Sci Total Environ. 2008;396(1):42–51. doi: 10.1016/j.scitotenv.2008.01.065. [DOI] [PubMed] [Google Scholar]
  21. Handel AE, Williamson AJ, Disanto G, Dobson R, Giovannoni G, Ramagopalan SV. Smoking and multiple sclerosis: an updated meta-analysis. PLoS One. 2011;6(1):e16149. doi: 10.1371/journal.pone.0016149. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Harding K, Tilling K, MacIver C, Willis M, Joseph F, Ingram G, et al. Seasonal variation in multiple sclerosis relapse. J Neurol. 2017 Avr; doi: 10.1007/s00415-017-8485-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Havard S, Deguen S, Zmirou-Navier D, Schillinger C, Bard D. Traffic-related air pollution and socioeconomic status: a spatial autocorrelation study to assess environmental equity on a small-area scale. Epidemiology Camb Mass. 2009;20(2):223–230. doi: 10.1097/EDE.0b013e31819464e1. [DOI] [PubMed] [Google Scholar]
  24. Heydarpour P, Amini H, Khoshkish S, Seidkhani H, Sahraian MA, Yunesian M. Potential impact of air pollution on multiple sclerosis in Tehran, Iran. Neuroepidemiology. 2014;43(3–4):233–238. doi: 10.1159/000368553. [DOI] [PubMed] [Google Scholar]
  25. Janes H, Sheppard L, Lumley T. Case-crossover analyses of air pollution exposure data: referent selection strategies and their implications for bias. Epidemiology Camb Mass. 2005;16(6):717–726. doi: 10.1097/01.ede.0000181315.18836.9d. [DOI] [PubMed] [Google Scholar]
  26. Javierre BM, Hernando H, Ballestar E. Environmental triggers and epigenetic deregulation in autoimmune disease. Discov Med. 2011;12(67):535–545. [PubMed] [Google Scholar]
  27. Jin Y, de Pedro-Cuesta J, Söderström M, Stawiarz L, Link H. Seasonal patterns in optic neuritis and multiple sclerosis: a meta-analysis. J Neurol Sci. 2000;181(1–2):56–64. doi: 10.1016/s0022-510x(00)00408-1. [DOI] [PubMed] [Google Scholar]
  28. Kalincik T. Multiple sclerosis relapses: epidemiology, outcomes and management. A systematic review. Neuroepidemiology. 2015;44(4):199–214. doi: 10.1159/000382130. [DOI] [PubMed] [Google Scholar]
  29. Kobelt G, Texier-Richard B, Lindgren P. The long-term cost of multiple sclerosis in France and potential changes with disease-modifying interventions. Mult Scler Houndmills Basingstoke Engl. 2009;15(6):741–751. doi: 10.1177/1352458509102771. [DOI] [PubMed] [Google Scholar]
  30. Lauer K. Environmental risk factors in multiple sclerosis. Expert Rev Neurother. 2010;10(3):421–440. doi: 10.1586/ern.10.7. [DOI] [PubMed] [Google Scholar]
  31. Laurent O, Pedrono G, Segala C, Filleul L, Havard S, Deguen S, et al. Air pollution, asthma attacks, and socioeconomic deprivation: a small-area case-crossover study. Am J Epidemiol. 2008;168(1):58–65. doi: 10.1093/aje/kwn087. [DOI] [PubMed] [Google Scholar]
  32. Maclure M. The case-crossover design: a method for studying transient effects on the risk of acute events. Am J Epidemiol. 1991;133(2):144–153. doi: 10.1093/oxfordjournals.aje.a115853. [DOI] [PubMed] [Google Scholar]
  33. Marcazzan G, Vaccaro S, Valli G, Vecchi R. Characterisation of PM10 and PM2.5 particulate matter in the ambient air of Milan (Italy) Atmos Environ. 2001;35(27):4639–4650. [Google Scholar]
  34. McKay KA, Jahanfar S, Duggan T, Tkachuk S, Tremlett H. Factors associated with onset, relapses or progression in multiple sclerosis: a systematic review. Neurotoxicology. 2016 Avr; doi: 10.1016/j.neuro.2016.03.020. [DOI] [PubMed] [Google Scholar]
  35. Mehrpour M, Shams-Hosseini NS, Rezaali S, Sahraiian MA, Taki S. Effect of air pollutant markers on multiple sclerosis relapses. Iran J Public Health. 2013;42(10):1167–1173. [PMC free article] [PubMed] [Google Scholar]
  36. O’brien R. A caution regarding rules of thumb for variance inflation factors. Qual Quant Int J Methodol. 2007;41(5):673–690. [Google Scholar]
  37. Ogawa G, Mochizuki H, Kanzaki M, Kaida K, Motoyoshi K, Kamakura K. Seasonal variation of multiple sclerosis exacerbations in Japan. Neurol Sci J Ital Neurol Soc Ital Soc Clin Neurophysiol. 2004;24(6):417–419. doi: 10.1007/s10072-003-0200-5. [DOI] [PubMed] [Google Scholar]
  38. Oikonen M, Laaksonen M, Laippala P, Oksaranta O, Lilius EM, Lindgren S, et al. Ambient air quality and occurrence of multiple sclerosis relapse. Neuroepidemiology. 2003;22(1):95–99. doi: 10.1159/000067108. [DOI] [PubMed] [Google Scholar]
  39. R Core Team. R: A Language and Environment for Statistical Computing [Internet] R Foundation for Statistical Computing; Vienna, Austria (Disponible sur): 2014. < http://www.R-project.org/>. [Google Scholar]
  40. Roux J, Bard D, Le Pabic E, Segala C, Reis J, Ongagna JC, et al. Air pollution by particulate matter PM10 may trigger multiple sclerosis relapses. Environ Res. 2017;156:404–410. doi: 10.1016/j.envres.2017.03.049. [DOI] [PubMed] [Google Scholar]
  41. Salvi F, Bartolomei I, Smolensky MH, Lorusso A, Barbarossa E, Malagoni AM, et al. A seasonal periodicity in relapses of multiple sclerosis? A single-center, population-based, preliminary study conducted in Bologna, Italy. BMC Neurol. 2010;10:105. doi: 10.1186/1471-2377-10-105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Sawcer S, Franklin RJM, Ban M. Multiple sclerosis genetics. Lancet Neurol. 2014;13(7):700–709. doi: 10.1016/S1474-4422(14)70041-9. [DOI] [PubMed] [Google Scholar]
  43. Spelman T, Gray O, Trojano M, Petersen T, Izquierdo G, Lugaresi A, et al. Seasonal variation of relapse rate in multiple sclerosis is latitude dependent. Ann Neurol. 2014;76(6):880–890. doi: 10.1002/ana.24287. [DOI] [PubMed] [Google Scholar]
  44. Thibaudon M. Report of the activity of RNSA (National aero-biological surveillance network), 2003. Eur Ann Allergy Clin Immunol. 2004;36(7):252–255. [PubMed] [Google Scholar]
  45. Tremlett H, van der Mei IAF, Pittas F, Blizzard L, Paley G, Mesaros D, et al. Monthly ambient sunlight, infections and relapse rates in multiple sclerosis. Neuroepidemiology. 2008;31(4):271–279. doi: 10.1159/000166602. [DOI] [PubMed] [Google Scholar]
  46. Vojinović S, Savić D, Lukić S, Savić L, Vojinović J. Disease relapses in multiple sclerosis can be influenced by air pollution and climate seasonal conditions. Vojnosanit Pregl. 2015;72(1):44–49. doi: 10.2298/vsp140121030v. [DOI] [PubMed] [Google Scholar]
  47. Watad A, Azrielant S, Soriano A, Bracco D, Abu Much A, Amital H. Association between seasonal factors and multiple sclerosis. Eur J Epidemiol. 2016;31(11):1081–1089. doi: 10.1007/s10654-016-0165-3. [DOI] [PubMed] [Google Scholar]
  48. Yoo W, Mayberry R, Bae S, Singh K, (Peter) He Q, Lillard JW. A study of effects of multicollinearity in the multivariable analysis. Int J Appl Sci Technol. 2014;4(5):9–19. [PMC free article] [PubMed] [Google Scholar]

RESOURCES