Spt an Spatiotemporal Fpidemilogy 34 (2020) 100358 Contents lists available at ScjonceDirect Spatial and Spatio-temporal Epidemiology journal homepage: wwvw.olsoviorcomvlocatolsste ELSEVIER Exploring spatio-temporal patterns in mental health related ® emergency department use from children and adolescents Michelle Thiessen, Qj Cui’, X. Joan Hu’, Rhonda J. Rosychuk®* * Deparment of Seite and Actuarit Scece, Sion Fer Unter, Bumaby rh Clumbig Canada ® eparmet of Pediatrics, Univers of Aber, Edmonton, Aber, Canada wipe. Maneba, Cardo ARTICLE INFO ABSTRACT ‘ace on Receicd 29 Nowembr 2019 Revie 15 Jone 2020, ‘Accepted 18 Je 2000 alae anne 20 ne 2020 ‘To understand the spatio-temporal patterns and associated isk factors with the frequency, we analyze records of mental health related emergency department (MHED) visits ftom youth, The data ate extracted for the period 2002-2011 trom the population-based, provincial health administrative data systems of Al- bora, Cada. Guided by a descriptive analysis, we conduct generalized linear regression analyses ofthe counts of MHED visits from various health areas. Seasonal elects are examined via thee different types Teyworts: of functions, incling trigonometric functions. We specify the temporal cortelation using an avtoregres> Generated nest model sive model of order 1 and formulate the spatial correlation by a random effects model, Our analysis re~ Mined ees vealsa strong seasonal pater and indicates thatthe MHED visit counts ae significantly associated with ereson anys ‘ge, gender, and 2 proxy for socio-economic statu. The final statistical model may be used to forecast Seasonal eet Future MHED use and identify regions and groups at 2 higher risk tothe MHEDs. {© 2020 Elsevier Ld. Al rights reserved 1. Introduction by children and youth aged less than 18 years old for substance Mental health is a relatively understudied subject matter, yet itis estimated that 29% of Canadian youth aged 13-19 years are affected by mental health issues (Mental Health Commission of Canada, 2012). Unfortunately only 20% of those who need men- tal health services receive them (Leitch, 2007). There is a lack of ‘community-based supports for mental health services, and inpa- tient and outpatient treatment options in Alberta, Canada. This of- ten leads to families seeking for help at emergency departments during crises (Newton et al, 2011), Mental health related emer sgeney department (MHED) visits can often be avoided if treatment is sought out before the crisis point is reached. Learning more about when and where the capacity of the pediatric mental health care system is exceeded by the need can help the province of Al- berta understand where more resources are needed and reduce ‘crowding in emergency departments. Potential policy implications partly motivated the research presented in this paper. Researchers have previously examined geographical clustering fof MHED visits. Mariathas and Rosychuk (2015) studied three different spatial cluster detection methods that differed in their choice of distributional assumption. Emergency department visits * conresponing autor ‘mal eddies: rhonda osyehukuatberac (R Rosy) ups:oLorg/as016yste.2020100358, 1977-58450 2020 Eee LAI rights reserve use during April 1, 2007 to March 31, 2008 were used. Statist- cally significant clusters were found in northern Alberta, parts of the Edmonton region, and southwestern Alberta for all three de- tection methods. Rosychik et al (2014) used emergency depart ment visits due to a mood disorder for Albertans aged 10-17, The Kulldorff-Nagarwalla (KN) spatial sean test (Kulidorff and Nagar ‘walla, 1995) foune! three potential clusters over space and time in ‘the majority of northern Alberta between 2007 and 2011, in a sin ale sub-regional health authority (sRHA) in the southwest part of ‘the Central Zone between 2005 and 2008, and another in the Cen- tral Zone between 2008 and 2011. Rosychuk etal. (2016) used the KN spatial scan on Albertans aged 15-17 years during 2002-2011 ‘who visited an emergency department for self-harm and had no physician follow-up visit within 14 days post-ED visit. They found 4 cluster in northern Alberta from 2002-2006 andl in southern AI berta between 200-2007, Marginal regression models have also been used to exam- ine MHED visits. Hu and Rosychuk (2016) and Rosychuk et al. (2018) investigate age-varying effects of risk factors on pediatric MHED visit frequencies by marginal regression analyses of the data. They show that older male subjects have lower MHED visit Frequencies compared to females than younger males. Their stl ies assume that the MHEDs were generated from independent sub- jects. The MHED visits are in fact naturally clustered according to the associated health sub-regions in Alberta, and itis important to2 [M.Missen, QC and XH t/a an Spario-emport Epcot 34 (2020) 10058 know about the correlation. In addition, a good understanding of seasonal effects to MHED use is often desirable, We aim to explore the spatio-temporal patterns in the MHED use and identify important risk factors associated with the frequency. The rest of this paper is organized as follows. Section 2 starts with a descriptive analysis of the PMHC dataset. ‘The generalized linear mixed effects model for the final inferential analysis emerges from analyses of the data under various plausible regression models. Section 3 describes the model associated esti- ‘mation procedure and presents the analysis results with the PMHC dataset, Final remarks and future work are given in Section 4, 2. PMHC dataset and statistical modelling 2, Alberta pediatric mental heath (PMHC) dacasee ‘The PMHC dataset used in this paper was taken from two population-based administrative databases in Alberta, the Ambula- tory Care Clasifieation System (ACCS) and the Population Registry File (PRF), The ACCS database provides Alberta MHED visit infor- ‘mation. The demographic and geographic data ate from the PRF database. The individuals of interest are Alberta residents who had at least one MED visit and were younger than 18 years old at the time of their MHED visit during the observation window April 1, 2002 to March 31, 2011. An Alberta resident is defined as an Individual who is registered in the Alberta Health Care Insurance Plan (AHCIP). Geographic location data are reported as the resi- dence at fiscal year end and thus our calculations use fiscal years rather than calendar years. In 2003, Alberta was divided into 9 regional health authorities (RHAs), and further divided into 70 sub-regional health authorities (sRHAs) Fig. 1 shows the choropleth maps of the number of MHED Visits per 1000 sRHA population over 9 fiscal years, Five intervals Were chosen based on Fisher Jenks natural breaks algorithm (Jenks, 1977) as they provided good spread of the rates of MHED vis- its and bad both the highest goodness of vatiance fit measure and tabular accuracy index compared to other approaches considered. We can see that sRHAs in RHA 3 and RHA 6, the major urban areas of the Calgary and Edmonton area, respectively, are in the higher range of MHED visits per 1000 sRHA population. Interestingly, the sRHAs in RHA 8, the northern area of the province, isin the lower range of MHED visits in the early 2000s but has an increase in visit rates and is now in the higher range bracket for 2010/2011. These choropleth maps show that the rate of MHED visits is different not only across sRHAs, but over time as wel Fig. 2 shows the monthly variation of MHED visits for the nine ifferent RHAs. The RHAs appear to follow a similar seasonal pat- tern. They all have a decrease in MHED visits in the summer ‘months, July and August, and increases in the winter months. Oc- tober and November, as well as March and May. This motivates ex- ploring a cyclic seasonal effect for the MHED visits. ‘The PMC dataset has the following demographic information collected on the individuals, The individual's age at the time of the IMHED visit and at fiscal year end, the proxy for socio-economic status (pSES), and gender. The age of the individual at fiscal year fend means that some individuals who were aged 17 at the time ‘of MHED visit might have turned 18 by the end of the fiscal yea. ‘The population data records the age at fiscal year end. In order to match the population data, the age at fiscal year end is cho- sen, The proxy for socio-economic status included three categories based on level of income support or source of payment for pre- ‘mium, denoted as pSES}, pSES;, and pSES;, ‘Table 1 summarizes the number of MHED visits by demo- graphic information with data taken from the ACCS and PRF databases. From Table I, it can be seen that females, teenagers aged 13-18 at the fiscal year end and pSES; individuals comprise of the majority of MHED visits for all of the 9 fiscal years. The age groups O-5 and 6-12 years old are combined in the analysis due to low counts of MHED visits 22. Notation and model specifications This paper focuses on the primary response Yt 1: x) to be the number of the MHED visits at time ¢ and region 7 with co- variates x. Specifically, the region index r is 1,...,70 for the 70, SRHAs, We use R to denote the RHAS, where R— 1....9 for the 9 RHAs. The number of sRHAS within RHAS varies by RHA, We consider the time starting at April 1, 2002 and with the unit of 228 day block. Thus each fiseal year has 13 time units (blocks) and 117 with the MHED data. The advantage of using 13 blocks over a monthly time unit is that each time unit has an equal amount of days. This ensures the variations in counts be- tween blocks will not be due to differences in amount of days. An- other method would have been to introduce an offset in the Pois- son regression that adjusts for the unequal number of days in the ‘months using (365.25/12)/(days in month ¢) (Barnett and Dobson, 20100. Since the response variable is in the form of counts, we as- sume it follows a Poisson distribution conditional on its covar ates, The covariate vector # includes as its components the vari ables of age group, gender, and pSES. We use X = (I.xi.§a.%3, 4)" Where the age group indicator xy =1 for the group of age 13-17 year-old, gender variable xp =1 for male, and x3 and xy are the dummy variables of the pSES variable with x3 = 1 and xq= 1 for the second and third pSES categories, respectively. Here the base- line references for the three variables are age 0-12, female, and the first pSES category. Let S(t; 1) be the size of population of Al- beertans in sRHA r aged 0-18 at time «The real population size at time F and region r is mostly unknown as itis only recorded for the SRHA level at the fiscal year end. For the purposes of this paper, we assume the population remains constant throughout the fiscal year, This assumption is not expected to have a large impact tn the Poisson regression model when the population size is rela- tively large and the fluctuations throughout the year are expected to be minor. Let Y(t) follow an autoregressive model of order 1 (Le, AR(1)) and 1(7) be a spatial random effect. We use R'S™® to denote the RHA 3, RG, RY oF the other RHAS. We consider the final model emerging from the discussions above and in Sections 2.21 and 2.2.2 below as follows: tos[slresr9p.b0.9,5e.7]] = Bovey vt Rem + (6b 0+ epee (ene 2) (ernsosin(222) Bas Ass + Prarixg + epet + one + Y(t), q) Where @p is the intercept, @), i= 1,...,5 are fixed effects and by(r) and a(r) are random effects of the demographic factors, and 6. A. and B are fixed seasonal effects and time trend, and bg(1), b(t) and Bp(r) are random seasonal effects and time trend. We will present further the structure of y(t) and (7 in Section 2.2.2 Inthe subsequent sections, we will explore the model and show evidence to justify the specific terms in Model (1). Model (1) is a special case of the following conventional generalized linear mixed effects model: tog [ELV 7.x), 006.7, S1e.9]] =XBE,7) + Aste, rlogste 7) 4+ Broxit + Brxina + Bratixe + Bans F Breeats FOG! bonxat + YO) Emr) — (2)[Messen Cu nd XY. He Spi ant Spier! Epldemiaay 34 (2020) 10388 3 (a) (e) (f) pana pene st po a sat pap (g) (h) @ ig. 1. MHED Wists per 1000 sRIUN in Ascal yeas: Fg (2) 200272003; Fs. 1) 20032004: Fie) 200472005; Pe 1¢) 20052006; Rs He) 2006)2000: Fa 10 2007/2008 1c) 2005/20: fg, (20087200, i 18) 200701.[M.Missen, QC and XH t/a an Spario-emport Epcot 34 (2020) 10058 8 3 os- < ar g \ a © oe. fay 3 Si Se EAS 5 Sean oe? 2 +. om 8 on 58 i Rs a ye a rr ar Month 2, Nano MED wis e100 RA pti oer mon in eee on sl ye with ll the ewo-fctor interactions. assumed 10 be con- + Example I. Asume a linear time tend and assign a eoetficient stant across time and space, and B(¢,1) = (Pott, 1). Bu(t,P). Ba(t.1).Pa(C.1), Balt, DY to be Biter) = att, r)-+ btF), Where aj(t.r), i=0,....5 are fixed effects and the ran- dom effects by(t,r), 1=0,...,5 with (1) = (ba(,1),D1(0), 6.1), ba (6.1). Ba(E.7).Bs(E. 1)" ~ MVNO, E16.) 221, Fixed effets specifications The fixed effect in the intercept a(t r) can be expressed as, ap (tr) = 8p(0) + W(C,7), where M(t, 7) describes the seasonal effect and time trend. ‘One may begin with considering the seasonal effect and time trend each of the regions: for region UE) = Serena (E19 + Foeasonat.sie) (6. 1), vriere J Tews 13 for the 13 blocks of size 28 days in a fis cal year and ¢ 117 for the 13 x 9 time points, and firenge sd Frama, 2 soothing functions for the ime end ted the cle seasonal pattern, respectively: The smoothing functions ‘Femoral ANd Sends CaN be represented using a cyclic smoothing “pine in tin plate smoothing spline, especie. For example the ei eUbie SMe I Fa) = 211) the basis functions Bj. A characteristic oF the cyte cube spine is Seamer yaaa rep (Woo S500 We consider particulary the following two examples forthe seasonal effect and time ted for each of the 12 = 13 ~ 1 blocks of size 28 days EEF) = GC IE+ B/C"), where j 13 for the 13 blocks of size 28 days in a year, and block 1 is the reference group g) =0. + Example T2, Assume a linear time trend and a seasonal pattern that is deseribed by trigonometric functions: me) 3 B ‘The cyclic cubie smoothing spline for fgaamaye) CAM be Seen as an extension of Example TI: the spline smoothes over the dis cretized seasonal effect in the example. The thin plate spline may bee used to check for whether a linear time trend is adequate or whether a polynomial time trend is required, Fig. 2 shows the estimated time trend and the seasonal ef- fect with the 13-1=12 components in the model of Example Ti, and with the trigonometric model of Example T2 for each RHA and the Alberta average. The seasonal patterns ate similar in the models of Examples TI and T2. To capture the seasonality in the MHED counts, the trigonometric model T2 is recommended over the cyclic cubic spline model and the 131 factors for the blocks of size 28 days. This is due to the trigonometric model hhaving fewer parameters compated to the seasonal factors model and trigonometric model being more interpretable compared ta smoothing splines. The expressions cos (27/13) + sin (2t/13) ac- count for the seasonal variation over time with the coefficients A(t nies) 10 years were analyzed using. Bayesian hierar chical models and Moran's I was also used to identify clustering in local areas (Lin et al.. 2019). During 2005-2014 for eight coun- ties in New York state.Zhang et al, (2020) looked at ED visits for patients of any age for several health conditions, including mental Gisorders, with an emphasis on comparing the time period of Su- perstorm Sandy with the same period in other years. The authors considered spatial clustering to group census tracts and developed ronlinear models to evaluate health risk and develop a community vulnerability index. These study results are not directly comparable to our study given that the jurisdictions, age group, diagnosis, and ‘methodological approaches vary considerably Overall, spatio-temporal patterns in MHED use is evidenced by our analysis, and our model helps to identify important demo. graphic factors from gender, age, and socio-economic status. The Analysis reveals that some demographic groups are more suscepti- bie to experience MHED visits than the other groups. A variety of issues need to be addressed to further enhance the value of this research, Our model assumes lag-1 temporal autocorrelation, The est imate of p is found to be significant from zero. A larger lag might show significant temporal autocorrelation for time points further apart. The analysis outcomes show that sRHAs in the same RHA and adjacent sRHAs are not significantly correlated. This suggests that the hierarchical structure of sRHAS in RHAS is not very strong ‘The spatial weights matrix considered in this paper is based on the adjacency of two regions, which is based on the Cliff and Ord (1973) spatial autocorrelation definition. Different forms of spa- tial contiguity could be considered, such as distance thresholds Distance-based spatial contiguity may yield different results ‘AS pointed out by Rue et al. (2009), the parameter esti- ‘mates can be sensitive to the chosen priors. To explore the ef fect of the prior distribution on the mode! and parameter esti ‘mates, we assign weakly informative prior distributions to the vari- ance parameters. The prior distributions for p and iy remain the same but the remaining parameters are set such that —log(a?) ~ log Gamma(0.1.0.1). The parameter estimates with weakly infor ‘mative prior distributions are similar aside from the estimate for the RHA random effect and the random slope for the time trend grouped by region. The highly vague prior distributions estimated these effects to be very small but the weakly informative prior dis- tributions estimated them to be lager due toa stricter distribution setting, Despite the differences in these parameter estimates, the information criterions remain largely the same. Moreover, individuals in this data can have multiple MHED visits which makes the data recurrent event data, meaning that MHED visits made by the same individual are correlated. This pa- per focuses on aggregate-level data but future investigations could incorporate this correlation structure by considering individual- level data, In addition, the spatial and temporal effects are[Messen Cu nd XY. He Spi ant Spier! Epldemiaay 34 (2020) 10388 " 06 WHED vets per 100 RA popaltin of en femal RPP fe pe i i gs —— 01 6 pr 0 th pon tn es 5 i fs r a s one wt pr st opstan en a eon spond 8 dg i i: a be a a 3 Fig. 7 Pos of tue and fed NED visits per 1000 sRHA population for females aged greater chan 12 in sRHA RAO with socio-economic satus SES, SES; oF BE, THe lack ine 1 the ru rates. the Bie ine i the Red posterior means andthe dashed fed lines ae the 25% and 67.5% quantiles ofthe ed posierie considered separately in the model. A more informative: model 3 inchide an interaction term for space and time. This can ‘be implemented through a random effect that takes the Kronecker product of 2 temporal and spatial structure matrix, such as the AR(1) and CAR structure (Clayton, 1996), ‘We remark that there are limitations with the PMHC dataset. First, the population count information is incomplete since it is only available at the fiscal year end. Interpolation could be used to estimate the population counts in between years. Second, the records of the MHED visits were from individuals less than 18 years of age during April 1, 2002, to March 31, 2011, which gives Fise to doubly censored data. Third, the smallest unit for the spa- tial information regarding the subject's place of residence is only available at the sRHA level, The spatial information is unavailable as point data, but rather polygon data, so there is a loss of in formation. Fourth, the presence of other mental health services in ‘the region can be a potential confounder. Some regions could have Fewer MHED visits due to access to other mental health services2 [M.Missen, QC and XH t/a an Spario-emport Epsom 34 (220) 10058 However, the PMHC dataset does not contain information on whether individuals visited other mental health services. Future investigations plan on including information such as the patient’ triage level and diagnosis. Triage level gives an indication of the severity ofthe child or youth’s MHED visit, Lastly, knowing. Which sRHAs are expected to experience large amounts of MHED Visits given their population size, and at what time petiod to ex- pect these MHED visits could be helpful to the government of Al- berta. In the future, we plan on investigating the predictive abili ties of our model. Declaration of Competing Interest None, Disclaimer ‘This article is based in part on data provided by Alberta Health, ‘The interpretation and conclusions are contained herein ate those lof the researchers and do not necessaily represent the views of the Government of Alberta, Neither the government nor Alberta Health expresses any opinion in relation to this study. CRediT authorship contribution statement Michelle Thiessen: Conceptualization, Formal analysis, Investi- gation, Methodology. Software, Validation, Visualization, Writing - original draft. Qi Cui: Investigation, Validation, Visualization, Weit- ing ~ original draft, Writing ~ review & editing, X. Joan Hus Con- ceptualization, Funding acquisition, Methodology, Project adminis- tration, Supervision, Writing - original draft, Weiting ~ review & editing, Rhonda J. Rosychuk: Data curation, Funding. acquisition, “Methodology, Project administration, Supervision, Writing - review 8 editing Funding ‘This research is supported by research grants to XJH_and IR from the Canadian Institutes of Health Research (CIHR) and the Natural Sciences and Engineering Research Council of Canada (NSERC). The sponsors had no role in the study design, in the col- lection, analysis and interpretation of data, in the writing of the report, and in the decision to submit the article for publication, Acknowledgements ‘The authors thank Alberts Health for facilitating access to the data References ‘Ares A. 2013. Catericl Data Analysis hid ed, Wey Series in Probability and ‘attics Barnett AG, Dobson, Al, 2010, AnasingSeasonat Health Data, Sige. Besa, NE. Clayton, BG, 1895. Apprnimate inference in generated linear ‘ined model Am, Stat, As. 88 (42) 9-25. Cayton, DG. 196, Generaed near mised modes: Gis, R, Richardson, S. ‘spegtlter, DJ (Eds. Maroy Chan Monte Caro in Practee. Chapman and Halt Londen, pp 275-301 Cl, AD. Od. 1873, spatial Auteoretation, Fon, xa, Cy Ging, Hedy 2008 Predictive model asessment fr count data, Biometrics 6 (4) 1254-1061, Fanshawe, Ek, Dig Pl, Rushton, , Sanderson, R, Lure, PWM. Ginna, SV, ere, MIS, Parker. Craton, Pseuiol, 200K Modeling spa° ‘Svenoral ation a pase parce mate: 3 wot apps Fong. Rus, H. Wafl 210 Bayesian nerence for geeraie er mined, Tool isa (3 207-18, tu. Rosse, RJ. 2016, Margi ereson anlss of recutent events wih oaiencdcesing tes, imei 72, 13-1122 Jenks, Cf H7, Optimal Una Gsfenton fr Chorosth Map, Tcl Repr. Deparient of Geogr Ulery of Kansas, amen aL, Sapo, EK Bara ED. Holga. Kee A ile, A208 ends ia even emergency apotent ts sing youth and 9oung at i the Ue Pei 143 4), 0.542 pes 2018- 212, ‘uot Me Nagrwale Ne 105 Spt! dea str: detection and inference Sau Me (8), 78-830, Leth KX 2007 eating te Top: A Report By the AdMsr on Heat C- ren aa Yu. Heh Canad, Gta, ON esis Pace, RX 2008 ned Spal fonometcs CRC Press tin C¥. Bley, Gemems, C. Webb, Rt Cunt D. Hau, C2. Chang 5S, api, N. 2018 Spel patering and coats of selcharm in Manche, nla. Eide Pays, S27, a CH le JA Sh Li Sle, RM. 200, ike met ath Maal €< Back T- Doan, Q 2013 Trends in peat mercy departnent wi Tzaon for mesial healheated Wats) Peat 157 (505-90 Mariathas HH, Rony) 2015 An extn ft spat vent cuter rection methods SPS it. Gea tfom, 367-984, Mattos beret, WO, Les sil otels Ranta P2008. An autoeges- se approach to spatiotemporal dice mapping Sat Med 77, 263-238, Ncallodh CE. Sere, Si, Neahain, JM, 2008 Ceneralie, nc nd Mined ade second Jon Wey an Son, New York. Mette Common of nade 3012 angie dscns. caning ies! the meta Reaihstategy ior Cana Aaah at ps es tnenatheathcommsson fate sais State ENC econ Chang. tH Ling Q, Hastings. NS. 2015 Spaotemporet hare mode Sxleaniswee ke so aese Neco Gos, obs PF. 2013. Apt plson nace made! or expcng rarptie vartion i emergency deariment Wt}. Sa So. SA 7S fear, Newon, A. Rye. RJ. Al S.Cawthore, D. Caran, J. Dons, K Slam ‘echt 201 The Emergency Depart Compass res Mental Fea: Palate mena heath emergence in bert nad: Emergeny de artnet vst by dren apd yout ge 0 oT? Yeas, 2002-2008, Paton ton Aa Newton: AS. Save, K. Rosyehuk, RJ. 2016. Dees emergency deparment 88 fr lcohol nother dug se caste gopaphicaly? A population based reo Spectve cor st, Substance Use Mise 31 (9), 120-1244, ‘aot 1, 2008, Bayesian Noting Using WingBJCS. ann Wiley an Sons Diva 2012 Bayesian ana of contdonataregesive model. Aan Ih Staak 6 (1, 107138 Rsyehle RJ Johnson, DW. Uric, Done K, Newton, AS. 2016, Does emer en Spirent ise and ost-vst ysl care cluster grogtphly and ‘Eporay tor adlescnts wno shar? A population sed yea reo Specie coor stay fom Ata, Canada. BMC Peay 1, 29. Aoyehae RJ, Newton. AS. Hu, Ki 2018. Ap as he impact oF important xe iors on metal heath emereency deparnent vist Be Hes eres, Rosyel, RJ, Newton, AS. Ni X Uebuk L204 space and time usterng of Iblescenis emereney deparent sea posts pian cate fer Meed tisorders in Aber Canada» poplaonsased Sear respective sd. Cina Pb Heath 106 2) 108 fue, Martino, 8, Chopin, N, 2008, Approximate Bayesian inference for latent aia i ug sad id ape spots oh Ruzang, J. Bit, M., Ceci, E, Groene, G. Botte, A, Hargreaves, DS. Saxena, S. aire eaten clon aea le han 19 ye 4 popl Geonoased coho study incl fom 20070 2017 BA Open 10) ‘ab My Raye R210 Spatiotemporal modeling of saese mapping of ‘ain Canin | Su, 38 (4) 8-715, ean, QM. Lmbeh, LG, Sanderson, K,_ de Gra, B, Besa, My Trin, Ve Ticker El. Nel AL, 2008: Trend of emergency Sear presents ith meal eat goss by ap, usta 20045 2617 Secondary Una ants: Emergency Ne. Resta 91 () 10041072, Ver ota Peon EE ote Mth, Hanks, Fri 208 Spt teragtasve mot for taal trees fom Slog dt, ea Mong 5 3659. Wallet LA. Cain, 82, Xa Hl, Gland, AE, 107 Werarehatspti-empora mapping aese rates im Sat Asc $2 60)-67, Zhang. Kinny PR, DQ, Shean S- Rome, AOL Dong Stem, EX DZ, Xan, lawrence. WR, Ln 7 Hao, Yin 5.2020, How community ‘ley tacts jomiy fet mule est bucomes fer csasop he Some Eton ne 1341508,