> oqn'` (bjbj{P{P 4*::2222222hd *<>>>>>>$/
hb]2b2222<<22{w<0 y
y
y
2Hbb FZh222222Effectiveness of Ontarios Universal Influenza Immunization Program (UIIP)
Note: This description is based on Kwong JC, Stukel TA, Lim J, McGeer AJ, Upshur RE, et al. (2008) The effect of universal influenza immunization on mortality and health care use. PLoS Med 5: e211.
The impact of Ontarios UIIP on influenza-associated mortality, hospitalizations, and visits to emergency departments (EDs) and doctors offices was estimated using data from 1997 to 2004 (3 years before and 4 years after UIIP implementation). Influenza-associated outcomes were estimated by using multivariate regression models to generate baseline functions representing the hypothetical absence of influenza and then subtracting the expected baseline events from observed events during periods of influenza activity. The models controlled for age, sex, province, influenza surveillance data, and temporal trends.
Estimation of baseline function
First, separate Poisson regression models were ran for each outcome, according to province and age group. Event counts were aggregated by week and sex within province-age group combinations. The dependent variable was the weekly event count for males and females, and the offset parameter was the province-age group-sex population. Models controlled for sex; viral surveillance for influenza A, influenza B, and RSV; the seasonal percentage of A(H3N2) isolates; the percentage of mismatched strains; linear and quadratic terms to model annual trends; and sine and cosine terms with periods of 1 y to model seasonal fluctuations, as in previous studies. To account for fluctuations in health service delivery during Christmas and post-Christmas holiday periods, categorical terms were included in the health care use models for these time periods. The expected baseline was generated by setting the influenza-related variables in the model (i.e., weekly percentage of tests positive for influenza A and B, seasonal percentage of A[H3N2] isolates) to zero as these were the terms that tracked influenza season peaks. Variance overdispersion was incorporated in the estimates of all standard errors (SE) to account for clustering of outcomes within weekly strata since outcomes to individuals within these strata may not be independent.
Description of regression model
The multivariate Poisson regression model was expressed as follows:
ln(Y) = ln(population) + 0 + 1[sex] + 2[%FluA] + 3[%FluB] + 4[%RSV] + 5[%A(H3N2)] + 6[%mismatch] + 7[t] + 8[t2] + 9[sin(2t/52)] + 10[cos(2t/52)] +
Y represents the weekly number of events for a particular outcome (e.g., all-cause mortality) in a province for a specific age group and sex stratum. The offset term is the log of the annual province-, age- and sex-specific population size. 0 is the intercept and 1 estimates the effects of sex. 2 through 4 account for the weekly percentage of provincial specimens testing positive for influenza A, influenza B, and RSV, respectively. 5 accounts for the percentage of A(H3N2) isolates and 6 accounts for the percentage of circulating strains mismatched to vaccine strains in a season. 7 and 8 are the coefficients for the linear and quadratic time trend terms, with t expressed as the week since August 24, 1997 (1 to 416) divided by 52. 9 and 10 account for the seasonal cyclical pattern. We used a period of 1 year, as in previous studies. The error term represents random error in the model. Additional terms for Christmas holiday weeks and the post-Christmas holiday week were included in the health care use models to account for fluctuations in health care service delivery during holiday and post-holiday periods.
Weekly influenza-associated events were subsequently computed as the difference between the number of observed events and expected baseline events during periods of peak influenza activity, where expected counts were based on the adjusted Poisson models. These weekly estimates were aggregated to produce annual estimates. Overall and age-specific mean annual rates of influenza-associated outcomes were calculated for the periods before and after introduction of UIIP (pre-2000 versus post-2000), separately for Ontario and the other provinces combined.
Pre- and post-UIIP influenza-associated event rates were compared by dividing the adjusted postintervention rates by the preintervention rates to produce relative rates (RR) of UIIP effect separately for Ontario aJKRn {$&&'((h$7?hVh5hV6Uhm#hVhy~hVhVCJOJQJh.hVCJOJQJhjhVhVK {5#7%U((((gd.gd.dgd.gd.(nd the other provinces combined. The standard error of ln(RR) was computed assuming a Poisson distribution for the overall counts and incorporating the variance of the predicted baseline events. The pre-/post-RRs for Ontario and other provinces combined were compared using the z-test and expressed as a ratio.
Model fit was evaluated by examining the standardized Pearson residuals for outlying points and secular trends. We also evaluated the presence of influential provinces by removing them individually from the model and reestimating the UIIP effect. Although the model did not optimally fit the extreme, short-lived spikes that occurred during the peaks of the influenza season, the fit during the remainder of the season was reasonable. The Durbin-Watson d-statistic was used to test for autocorrelation in the residuals, both including and excluding these influenza season spikes, averaging the correlations across province- and age group-specific models. Analyses were performed using SAS 9.1 (SAS Institute). All statistical tests were computed at the 5% level of significance and were two sided.
Results of the regression analysis are presented in Table S6.
2P:pn/ =!n"n#n$n%SSN@N.Normald CJOJQJ_HaJmH sH tH X@X. Heading 2$<@&5;OJQJ\]^JaJN@N. Heading 3$@&5OJQJ\^JaJDA@DDefault Paragraph FontRiRTable Normal4
l4a(k@(No ListTOT.Heading 2 Char 5;CJOJQJ\]^JaJNON.Heading 3 Char5CJOJQJ\^JaJ2)@2.Page Number^J*K{&
&:&Y&o%&&oJK{5 Fq}0000(00(0000000000(((j m#.$7?n21~y~V5@vvxPvvXN N
NNN
F@4@H@UnknownGz Times New Roman5Symbol3&z Arial;"Helvetica7K@Cambria"h\\C'C'nnhh4d2HP$P.2JEFFECTIVENESS OF ONTARIO S UNIVERSAL INFLUENZA IMMUNIZATION PROGRAM (UIIP)Beate SandermbrownOh+'00$0D T`
LEFFECTIVENESS OF ONTARIOS UNIVERSAL INFLUENZA IMMUNIZATION PROGRAM (UIIP)Beate SanderNormal.dotmbrown2Microsoft Office Word@F#@$w@$wCGd}VT$m >&" WMFCW D6DVl'3UT#m EMFDV"'3\KhC
'3%Rp8@"Helvetica;"Helvetca?h@)0
20
L^0
dv%T{H
^@E@4L'3tEFFECTIVENESS OF ONTzzz88z8zTI
{-^@E@I
4L'3tARIO S UNIVERSAL INF8888z88zT.{^@E@.4L'3tLUENZA IMMUNIZATION {{888zz88T|{#^@E@4L'3\PROGRAM 8TpG*@E@L'3X(UIIP)C88CTTG)*@E@L'3P k nRpY@"Helvetica;"Helvetca?0<)0
20
L^0
dv%TpGm @E@L'3XNote: y].]..TTnG @E@nL'3PTcsfTG
@E@L'3his description is based on ]$T.]\TS8%].%\].%T.]\T]].\].T
G# @E@
KL'3Kwong JC, Stukel TA, Lim J, McGeer AJ, Upshur RE, et al. (2008) The effect .goy\\].Ty..o.]S]%.fo..]%.T..T]\8.oT..y\S]]8-yo..]..]%./8]\]]8.e]].]-.]T..TX 2 @E@ WL'3of universal influenza immunization on mortality and health care use. PLoS Med 5: e211.se]..]]%S]7T]%.$].%]]\T].%]]%S].%]].]]-]8.]%%.S.]]]-]]]%.].S]7].\T]..o]]o.\].]..]]]].TT3 @E@3 L'3P maYRp8@Times New RomanGz Times ew Roman?
)0
20
L^0
dv%TT
@E@z
L'3P ua[TDr@E@F)L'3The impact of Ontario s UIIP on influenzas zdY28dYY82dB2d8YC8dCN1CCo2dd28dB8dYdYYTTFr@E@FL'3P-s CTGr@E@GFL'3xassociated mortality,umXNNdY7Y8Yd2dC8Y877d2T2!r@E@F!L'3 hospitalizations, and visits to ti2ddNd88X88XY88ddN21Ydd2c8N87N28d1T^
">@E@_L'3emergency departments (EDs) and doctors offices was estimated using data from 1997 to 2004 (3 deYYCdYdYd2dYdYC8Yd8M2CzNC2Ydd2cdY8dCMC2dBB8YYN2YN2YN78Y8Yd2dM8dd2cY8Y2BCd2dded28d2dddd2Cd1T*S
@E@>L'3years before and 4 years after UIIP implementation). InfluenzadYYCN2dYBdCX2Ydd2d2dYYCN1YB8YC2CCo27d8YYd8Y88ddC22CdB8dYcYYTTT*
@E@TL'3P-tiCT0*$
@E@&L'3associated outcomes were estimated by YNNcY8X8Xd2dd8YdYN2YCY2YN78Y8Yd2dd2T@E@L'3using multivariate regressidqdN8dd2d888dYC8X8Y2BYdCYMN8T,"@E@GL'3on models to generate baseline functions representing the hypothetical dpdd2ddY8N28d2dYdYBY8Y1dYNY78dY2BddY88ddN1CYdBYMYd88dd18dY2cddd8dY78YY81T#@E@vcL'3absence of influenza and then subtracting the expected baseline events from observed events during haYdNYdYY2dB27dB8dYdYY2Ycd28dYd2Ndd7BYY88dd18dY2YcdYY8Yd2dXNY78dY2YdYc8N2BBe2ddNYCdYd2YdYd8N2ddB8dd2TP$n@E@BVL'3periods of influenza activity. The models controlled for age, sex, province, influenzadYC8ddN2dB28cB8d&" WMFC DDVYdYY2YY77d88d22zdY2ddY8N2Ydc8Cc88Yd2BdC2YdY21NYd22dCdd8dXY228dB8dYdYYT%#n@E@%BL'3t surveillance data, 1NdCdY788YcYY1dY8Y21TZI
:@E@L'3xand temporal trends. ioYdd28YddCY828BYddN22TTJ
Z
:@E@J
L'3P 63Z%TG@E@L'3Estimation of baseline functiontyoC8oC8zy8zC8zooo88yo8CzzoC8zzTTH@E@HL'3P mao%T#@E@ncL'3First, separate Poisson regression models were ran for each outcome, according to province and age \lo8CN822MYdYCX8Y2od8NNdd1CXdCYNN8dd2ddY8N2YCY1CXd2BdC2YYYd2cd8YdY22YYYdCd8dd28d2cBdd8dYY2Ydd2XdY2Tf@E@:CL'3group. Event counts were aggregated by week and sex within provincet dCddd22zdYd82Yddd8N2YBY2YddCYdX8Yc2dd2YYd2Ydd2NYd288d8c2dCdd8dYXTTf@E@:L'3P-lsCTPf@E@:L'3lage group combi \YcY2dCddd2Ydd8TQ^#f@E@Q:
L'3hnations. The \dY78ddN22ydY2TR;$2@E@fL'3dependent variable was the weekly event count for males and females, and the offset parameter was the dYdYddYd82cYC8Yc8Y2YN27dY2YYd8d2XdYd82Yddd82AdC2Y8YN2Ydd2BZY8YN22Ydd28dY2dBBNY82dYCYY8YC2YN28dY1T| l @E@ L'3\provincedCdd8dYXTTm @E@m L'3P-\lCT @E@ L'3`age groupsdYcY2dCdddTT
@E@ L'3P-idCT(
;" @E@
OL'3sex population. Models controlled for sex; viral surveillance for influenza A, rfNYd1dddd8Y78dd21ddY8N2Ydc7Cd87Yd2BdC2NXd82d8BY82NcCcY878YdYY2BdC27dB8dYdYY222TP!"@E@"+L'3influenza B, and RSV; the seasonal percentaSh8dB8dYdYY212Ydd2o828dY2NYYNddX81dYCYYc8YT!""@E@"5L'3ge of A(H3N2) isolates; the percentage of mismatched iddY2cB2CddC29Nd8Y7YN818dY1dYCYYc8YdY2cB38OY8YdYd2T#V$$@E@j$jL'3 strains; linear and quadratic terms to model annual trends; and sine and cosine terms with periods of 1 y N8CX8dN8188dXXC2Ydd2ddYdBX88Y18YCN28d2ddY82YdddY827CYddN81Ycd2N8dY2Ydd2XdN8dY27YCN288d2dYB8ddM2dB2d2d3T%)!b&@E@6&aL'3to model seasonal fluctuations, as in previous studies. To account for fluctuations in health ser\l8d2ddY82NYYNddY82B8dY8cY88ddN21YN28d1dCYd8ddN2M8cd8YN22zd2YYXddd82BdC2B8dY8dY78ddN28d2cYY88d1NYCTl*!%"b&@E@*!6&L'3Xvice stc8YX2TN'.(@E@("L'3delivery during Christmas and postdY88dXCd2ddB8dd2dC8N8YN2Ydd2ddN7TTN'.(@E@(L'3P-sdBTN'#.(@E@(BL'3Christmas holiday periods, categorical terms were included in the dC8N8YN2dd88dYd2dXC8cdN22YY8YdcC8XY828XCN2YCY28dY8dcYd27d28dY1T)X")@E@)bL'3health care use models for these time periods. The expected baseline was generated by setting the dYY88d1YYCY2cNY2ddY8N2BeC28dYNY188Y2dYC8ddN22zdX2YddYY8Yd2cXNY88cY2YN2dYdYCX8Yd2dd1NY878dd288&fWMFCDDVdY1T*+@E@+ L'3`influenza08dB8dYdYYTT*+@E@+L'3P-0BTH*+@E@+*L'3related variables in the model (i.e., weekCY7Y8Yd2dXC8Yc7YN28d28cY2edY82B82Y222XYdT*:$+@E@+7L'3ly percentage of tests positive for influenza A and B, 08d2dYBYYd7YdY2dB28YN8N1ddN888cY2BdC27dB8dYdYY22Ydd222T,-@E@f-ML'3seasonal percentage of A[H3N2] isolates) to zero as these were the terms that0NYYNddY82cYBYYd8YdY2dB2BdeB28Nd8Y8YNC18d2YXCc2YN28dYNY2XCY28dY18YCN28dY8T,#-@E@f-L'3 tracked influenza season 18CYYcYd18dB8dYdYY2NXYNdd2Tx~.^/@E@2/L'3\peaks. 0dYYdN22Tp~."^/@E@2/[L'3Variance overdispersion was incorporated in the estimates of all standard errors (SE) to ac0YC7XdYY2ddYCc8NcYCN8dd2YN17dYdCddCX8Yd18d28dY2XN88Y8YN2dB2Y881N8XddYCd2YBCdCN1CozC28d2YXTp"~.$^/@E@"2/L'3Xcount Ycdd82%636'36'6636'36'6636'36'6636'36'6636'36'6636'36'6636'36'6636'36'6636'36'6 6 36'36' 6
6
36'36'
6
636'36'6636'36'6
6
36'36'
6
636'36'6636'36'6636'36'6636'36'66~36'~36'66}36'}36'66|36'|36'66{36'{36'66z36'z36'66y36'y36'66x36'x36'66w36'w36'66v36'v36'66u36'u36'66t36't36'66s36's36'66r36'r36'66q36'q36'6 6 p36'p36' 6 !!6!o36'o36'!6!!""6"n36'n36'"6""##6#m36'm36'#6##$$6$l36'l36'$6$$%%6%k36'k36'%6%%."System-@"Helvetica- )2
dGEFFECTIVENESS OF ONT
)2
dARIOS UNIVERSAL INF
)2
dLUENZA IMMUNIZATION
2
d`PROGRAM
2
G(UIIP)
2
o @"Helvetica-2
GNote:
2
iT52
phis description is based on |2
KKwong JC, Stukel TA, Lim J, McGeer AJ, Upshur RE, et al. (2008) The effect a
2
GWof universal influenza immunization on mortality and health care use. PLoS Med 5: e211.
2
3 @Times New Roman-
2
G I2
G)The impact of Ontarios UIIP on influenzak
2
H-+2
Massociated mortality,=2
! hospitalizations, and visits to t2
G_emergency departments (EDs) and doctors offices was estimated using data from 1997 to 2004 (3
h2
1G>years before and 4 years after UIIP implementation). Influenza
2
1-D2
1&associated outcomes were estimated by 42
TGusing multivariate regressi v2
TGon models to generate baseline functions representing the hypothetical 2
vGcabsence of influenza and then subtracting the expected baseline events from observed events during h2
GVperiods of influenza activity. The models controlled for age, sex, province, influenza )2
E surveillance data, +2
Gand temporal trends.
2
-:2
GEstimation of baseline function
2
' -2
GcFirst, separate Poisson regression models were ran for each outcome, according to province and age h p2
2GCgroup. Event counts were aggregated by week and sex within provincec
2
2-"2
2age group combi2
2\
nations. The b 2
UGfdependent variable was the weekly event count for males and females, and the offset parameter was the 2
wGprovince
2
w|-2
w age group
2
w-2
wOsex population. Models controlled for sex; viral surveillance for influenza A,
L2
G+influenza B, and RSV; the seasonal percenta
[2
S5ge of A(H3N2) isolates; the percentage of mismatched
2
Gjstrains; linear and quadratic terms to model annual trends; and sine and cosine terms with periods of 1 y 2
Gato model seasonal fluctuations, as in previous studies. To account for fluctuations in health ser 2
vice n>2
G"delivery during Christmas and post
2
-n2
BChristmas holiday periods, categorical terms were included in the
2
$Gbhealth care use models for these time periods. The expected baseline was generated by setting the 2
GG influenza
2
G-J2
G*related variables in the model (i.e., week^2
Gy7ly percentage of tests positive for influenza A and B,
2
iGMseasonal percentage of A[H3N2] isolates) to zero as these were the terms that22
i tracked influenza season 2
Gpeaks. e2
q[Variance overdispersion was incorporated in the estimates of all standard errors (SE) to ac
2
count -՜.+,0hp
University of Toronto''KEFFECTIVENESS OF ONTARIOS UNIVERSAL INFLUENZA IMMUNIZATION PROGRAM (UIIP)O Effectiveness of Ontarios Universal Influenza Immunization Program (UIIP)( Estimation of baseline function( Description of regression modelTitle Headings
!"#$%'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdeghijklmpRoot Entry FۇwrData
1TableWordDocument4*SummaryInformation(&`DocumentSummaryInformation8fCompObjq
FMicrosoft Office Word Document
MSWordDocWord.Document.89q