%#############################################################
%#
%#   Program: summary_tables.nw
%#   Project: Traditional Healers: Knowledge and Treatment of
%#            HIV/AIDS in Zambézia, Mozambique
%# 
%#   PI: Carolyn M. Audet, PhD
%#   Statistician: Bryan Shephard, PhD
%#   Biostatistician/Programmer: Meridith Blevins, MS
%#   Purpose: Read in data for analysis.
%#   Notes: 
%#   Created: 23 November 2009
%#   Revisions: 
%#     
%#   
%#
%#############################################################
% setwd("/home/blevinml/Projects/MozHealers/Code"); Sweave("analysis-scripts-003.nw")

\documentclass[11pt]{article}
\usepackage[margin=0.75in]{geometry}
\usepackage{setspace,relsize}   % needed for latex(describe()), \code
\usepackage{moreverb}           % for verbatimtabinput
\usepackage{url}
\usepackage{amssymb}
\usepackage{color}
\usepackage[pdftex]{pdflscape}   % allows tables to be landscape when
\usepackage{longtable}  % for subject listings 
\usepackage{graphicx} 
\usepackage{graphicx, subfigure}
\usepackage{threeparttable}
\usepackage{rotating}
\newcommand{\HRule}{\rule{\linewidth}{0.5mm}}

\begin{document}
\SweaveOpts{prefix.string=graphics/plot} % Created a "graphics" subdirectory to save graph files in
\begin{titlepage}
  \begin{center}
    % Upper part of the page

    \textsc{\Large Statistical Report:}\\[0.5cm]
 
    % Title
    \HRule \\[0.4cm]
    { \Large \bfseries Use of allopathic and traditional health care providers\\ for HIV care in rural Mozambique:\\[8pt] HIV knowledge and health seeking behavior in Zamb\'ezia Province}\\[0.4cm]
    \HRule \\[1.5cm]

    % Biostatistician and Faculty Member
\begin{minipage}{0.4\textwidth}
\begin{center} \large
\emph{Biostatistician:}\\
Meridith Blevins, M.S.
\end{center}
\end{minipage}

 
\vfill
 
% Bottom of the page
{\large \today}
 
\end{center}
 
\end{titlepage}

% Hidden R code chunk --- reading in the data
<<echo = false, results = hide>>=
rm(list=ls()) # SHOULD BE FIRST LINE IN ALL R PROGRAMS - CLEARS NAMESPACE
setwd("/home/blevinml/Projects/MozHealers")
# read in study database
patients <- read.delim("Data/patients_20100217.csv", sep = ",")
patients$rural.urban <- tolower(patients$rural.urban)

# Call program to tabulate counts/proportions and median(q1,q3) overall and by HIV treatment
setwd("/home/blevinml/Projects")
source("table_functions.R")
setwd("/home/blevinml/Projects/MozHealers")
@

\section{Introduction}
Zamb\'ezia is currently experiencing one of the highest burdens of HIV in Mozambique.
Many people within the district still lack access to modern medicine, education about
 risk factors leading to HIV infection, or practical alternatives to risky behavior
 perpetuated through cultural norms and poverty. Traditional healers are highly
 regarded members of their communities who have a strong influence on the understanding
 of disease causation and treatment of their patients. There is a lack of understanding
 about particular practices and beliefs of traditional healers in Zamb\'ezia Province.
 Negative stereotypes held by Western-educated physicians coupled with the mistrust of
 healers toward doctors have created barriers to collaboration. In an effort to diffuse
 this situation, we performed a qualitative study of the medical beliefs and practices of
 traditional healers in six districts in Zamb\'ezia using semi-structured key informant
 interviews and an ethnographic study.


\subsection{Research Aims}
\subsubsection{To determine characteristics of patients who use traditional healers.}
\subsubsection{To determine characteristics of patients with varying levels of HIV knowledge.}

\section{Methods}
\subsection{Participants}
This is a cross-sectional study 349 Mozambican adults. 
 Currently, FGH clinics operate in six districts of Zamb\'ezia: Namacurra, Alto-Molocue, 
 Gile, Ile, Inhassunge, and Lugela. Healers are identified in each community through word
 of mouth; communities were identified at random in each district. Patients were solicited
 for interviews through home-to-home visits.
 Only patients older than 18 years old were interviewed. We interview
 Mozambican patients in an effort to identify trends in behavior, knowledge and treatments.
 Interviews were conducted within the communities, locations determined at the request
 of the interviewee, and all patients had the goals of the research explained to them
 before the interview begins. 

\subsection{Outcomes}

The primary outcomes among patient interviews will be the patient's use of a traditional healer and his/her knowledge of HIV/AIDS.

\subsection{Data Sources and Measurements}

The study contains 349 key informant interviews.
 Participants will be questioned about their experiences with traditional and
 modern medicine, health issues, views on the cause and treatment of particular
 illnesses, and preferences of medical practitioners. 

\subsection{Statistical Methods}

\noindent{\it 1. To determine characteristics of patients who use traditional healers.}\\
Summary characteristics will be tabulated by those patients who go to healers
 and those who did not.  A logistic regression may identify whether
 demographics or HIV knowledge are predictive of using a traditional healer.
 Specifically, age, sex, education level, location, HIV knowledge,
 and belief in HIV/AIDS cure have been identified as predictors of interest.
 If needed, missing data methods will be considered. Options include complete
 case analysis or multiple imputation.\\

\noindent{\it 2. To determine characteristics of patients with varying levels of HIV knowledge.}\\
Summary characteristics will be tabulated by patient's HIV knowledge.  HIV knowledge
 will be measured by the number of ways a participant correctly identifies a
 method of HIV transmission (range: 0-5). An ordinal (proportional-odds) logistic 
 regression may identify whether demographics or health-seeking behavior are predictive of
 HIV knowledge. Specifically, age, sex, education level, location, 
 condom use, and source of health information have been identified as predictors of interest.
 Additionally five separate logistic regressions using the same predictors may
 identify which demographics or health-seeking behavior are predictive of knowledge for
 each of five methods of transmission. 
 If needed, missing data methods will be considered. Options include complete
 case analysis or multiple imputation.\\

R-software 2.11.1 (www.r-project.org) will be used for data analyses. 


<<echo = false, results = hide>>=
# setwd("/home/blevinml/Projects/MozHealers")
source("Code/patient_table1.R")
source("Code/patient_table2.R")
source("Code/patient_table3.R")
source("Code/patient_table4.R")
source("Code/patient_table5.R")
library(Design)

set.seed(2009)

vcov.lrm <- vcov.default # THOMAS DUPONT FOUND AN ERROR IN FIT.MULT.IMPUTE THAT NEEDS THIS LINE FOR CORRECTION
patient.age <- patients$Age
patient.sex <- patients$Gender
patient.dt <- as.factor(patients$district)
patient.ru <- as.factor(patients$rural.urban)
patient.ed <- patients$Highest.level.of.Education
patient.cure <- as.factor(cureaids.rows[cureaids+1])
p <- data.frame(patients.visitTH, patient.age, patient.sex, patient.dt, patient.ru, patient.ed, patient.cure)
pp <- datadist(p) # Necessary in order to get summarization
options(datadist='pp') # Necessary in order to get summarization - defines ranges for all var.
set.seed(5)
p1.mi <- aregImpute(~ patients.visitTH + patient.age + patient.sex + patient.dt + patient.ed + patient.cure,n.impute=10,data=p)
# model3 <- lrm(patients.visitTH ~ patient.age + patient.sex + patient.dt + patient.ru + patient.ed + patient.cure, x=TRUE, y=TRUE)
# model3 <- fit.mult.impute(patients.visitTH ~ rcs(patient.age,3) + patient.sex + patient.dt + patient.ru + rcs(patient.ed,3), lrm, p1.mi, p)
# p=0.9061 evidence that patient.ed is nonlinear with log-odds.    
# p=0.0060 evidence that patient.age   is nonlinear with log-odds.  
model3 <- fit.mult.impute(patients.visitTH ~ rcs(patient.age,3) + patient.sex + patient.dt + patient.ed + patient.cure, lrm, p1.mi, p)
(getanova3 <- as.data.frame(anova(model3)))
model3.stats <- model3$stats
summary.model3 <- summary(model3,patient.age=c(25,35),patient.ed=c(3,4))
summary.model3a <- summary(model3,patient.age=c(25,45),patient.ed=c(3,4))

model3table1 <- rep(NA,14)
model3table1[1] <- paste("Age & & & &",myformat4(getanova3$P[1]))
model3table1[2] <- "\\\\hspace{.1in}25 (ref) & 1"
model3table1[3] <- paste("\\\\hspace{.1in}35", myformat3(summary.model3[2,4]), myformat3(summary.model3[2,6]), myformat3(summary.model3[2,7]),sep=" & ")
model3table1[4] <- paste("\\\\hspace{.1in}45", myformat3(summary.model3a[2,4]), myformat3(summary.model3a[2,6]), myformat3(summary.model3a[2,7]),sep=" & ")
model3table1[5] <- paste("Male", myformat3(summary.model3[6,4]), myformat3(summary.model3[6,6]), myformat3(summary.model3[6,7]), myformat4(getanova3$P[3]),sep=" & ")
model3table1[6] <- paste("District & & & &",myformat4(getanova3$P[4]))
model3table1[7] <- "\\\\hspace{.1in}Namacurra (ref) & 1"
model3table1[8] <- paste("\\\\hspace{.1in}Alto Molocue", myformat3(summary.model3[8,4]), myformat3(summary.model3[8,6]), myformat3(summary.model3[8,7]),sep=" & ")
model3table1[9] <- paste("\\\\hspace{.1in}Inhassuange", myformat3(summary.model3[10,4]), myformat3(summary.model3[10,6]), myformat3(summary.model3[10,7]),sep=" & ")
model3table1[10] <- paste("Education (per level)", myformat3(summary.model3[4,4]), myformat3(summary.model3[4,6]), myformat3(summary.model3[4,7]), myformat4(getanova3$P[5]),sep=" & ")
model3table1[11] <- paste("Can you cure AIDS? & & & &",myformat4(getanova3$P[6]))
model3table1[12] <- "\\\\hspace{.1in}No (ref) & 1"
model3table1[13] <- paste("\\\\hspace{.1in}Do not know", myformat3(summary.model3[12,4]), myformat3(summary.model3[12,6]), myformat3(summary.model3[12,7]),sep=" & ")
model3table1[14] <- paste("\\\\hspace{.1in}Yes", myformat3(summary.model3[14,4]), myformat3(summary.model3[14,6]), myformat3(summary.model3[14,7]),sep=" & ")

model3table <- paste(model3table1,"\\\\\\\\",sep=" ",collapse=" ")



# NEED TO CREATE MODEL 4 - 12/23

#Counts for Trans1: 301; 48
#Counts for Trans2: 130; 219
#Counts for Trans3: 5; 344
#Counts for Trans4: 20; 329
#Counts for Trans5: 68; 281
patient.trans1 <- as.factor(ifelse(!is.na(trans1) & trans1==1,1,0))
patient.trans2 <- as.factor(ifelse(!is.na(trans2) & trans2==1,1,0))
patient.trans5 <- as.factor(ifelse(!is.na(trans5) & trans5==1,1,0))
patient.trans <- as.factor(patients.trans)
usecondom[usecondom==1] <- NA
patient.usecondom <- as.factor(usecondom-2)
patient.aidsprob <- as.factor(patients$Was.aids.mentioned.as.a.most.common.ailment.)
patient.avoid <- rowSums(data.frame(avoid1,avoid2,avoid3,avoid4),na.rm=TRUE)
patient.avoid[patient.avoid==3] <- 2
patient.avoid1 <- ifelse(is.na(avoid1),0,avoid1)
patient.avoid2 <- ifelse(is.na(avoid2),0,avoid2)
patient.avoid3 <- ifelse(is.na(avoid3),0,avoid3)
patient.avoid4 <- ifelse(is.na(avoid4),0,avoid4)
p1 <- data.frame(patient.trans, patient.trans1, patient.trans2, patient.trans5, patient.age, patient.sex, patient.dt, patient.ru, patient.ed, patient.usecondom,patient.aidsprob, patient.avoid, patient.avoid1, patient.avoid2, patient.avoid3, patient.avoid4)

# and source of health information
set.seed(35)
pp1 <- datadist(p1) # Necessary in order to get summarization
options(datadist='pp1') # Necessary in order to get summarization - defines ranges for all var.
trans1.mi <- aregImpute(~patient.trans1 + patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob,n.impute=10,data=p1)
trans2.mi <- aregImpute(~patient.trans2 + patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob,n.impute=10,data=p1)
trans5.mi <- aregImpute(~patient.trans5 + patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob,n.impute=10,data=p1)
trans.mi  <- aregImpute(~patient.trans  + patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob,n.impute=10,data=p1)

# TRANS 1
# trans1.model4 <- lrm(patient.trans1 ~ patient.age + patient.sex + patient.dt + patient.ru + patient.ed + patient.avoid + usecondom, x=TRUE, y=TRUE)
# trans1.model4 <- fit.mult.impute(patient.trans1 ~ rcs(patient.age,3) + patient.sex + patient.dt + patient.ru + rcs(patient.ed,3) + patient.avoid + usecondom, lrm, trans1.mi, p1)
# p=0.6868 evidence that patient.age is nonlinear with log-odds.
# p=0.1493 evidence that patient.ed is nonlinear with log-odds.    
trans1.model4 <- fit.mult.impute(patient.trans1 ~ patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob, lrm, trans1.mi, p1)
(trans1.getanova4 <- as.data.frame(anova(trans1.model4)))
trans1.model4.stats <- trans1.model4$stats
trans1.summary.model4  <- summary(trans1.model4,patient.age=c(25,35),patient.ed=c(3,4),patient.usecondom=0)
# TRANS 2
# trans2.model4 <- lrm(patient.trans2 ~ patient.age + patient.sex + patient.dt + patient.ru + patient.ed + patient.avoid + usecondom, x=TRUE, y=TRUE)
# trans2.model4 <- fit.mult.impute(patient.trans2 ~ rcs(patient.age,3) + patient.sex + patient.dt + patient.ru + rcs(patient.ed,3) + patient.avoid + usecondom, lrm, trans2.mi, p1)
# p=0.7845 evidence that patient.age is nonlinear with log-odds.
# p=0.9721 evidence that patient.ed is nonlinear with log-odds.    
trans2.model4 <- fit.mult.impute(patient.trans2 ~ patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob, lrm, trans2.mi, p1)
(trans2.getanova4 <- as.data.frame(anova(trans2.model4)))
trans2.model4.stats <- trans2.model4$stats
trans2.summary.model4  <- summary(trans2.model4,patient.age=c(25,35),patient.ed=c(3,4),patient.usecondom=0)
# TRANS 5
# trans5.model4 <- lrm(patient.trans5 ~ patient.age + patient.sex + patient.dt + patient.ru + patient.ed + patient.avoid + usecondom, x=TRUE, y=TRUE)
# trans5.model4 <- fit.mult.impute(patient.trans5 ~ rcs(patient.age,3) + patient.sex + patient.dt + patient.ru + rcs(patient.ed,3) + patient.avoid + usecondom, lrm, trans5.mi, p1)
# p=0.9305 evidence that patient.age is nonlinear with log-odds.
# p=0.5043 evidence that patient.ed is nonlinear with log-odds.    
trans5.model4 <- fit.mult.impute(patient.trans5 ~ patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob, lrm, trans5.mi, p1)
(trans5.getanova4 <- as.data.frame(anova(trans5.model4)))
trans5.model4.stats <- trans5.model4$stats
trans5.summary.model4  <- summary(trans5.model4,patient.age=c(25,35),patient.ed=c(3,4),patient.usecondom=0)
# NO TRANS 
# trans.model4 <- lrm(patient.trans ~ patient.age + patient.sex + patient.dt + patient.ru + patient.ed + patient.avoid + usecondom, x=TRUE, y=TRUE)
# trans.model4 <- fit.mult.impute(patient.trans ~ rcs(patient.age,3) + patient.sex + patient.dt + patient.ru + rcs(patient.ed,3) + patient.avoid + usecondom, lrm, trans.mi, p1)
# p=0.6306 evidence that patient.age is nonlinear with log-odds.
# p=0.7408 evidence that patient.ed is nonlinear with log-odds.    
trans.model4 <- fit.mult.impute(patient.trans ~ patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob, lrm, trans.mi, p1)
(trans.getanova4 <- as.data.frame(anova(trans.model4)))
trans.model4.stats <- trans.model4$stats
trans.summary.model4  <- summary(trans.model4,patient.age=c(25,35),patient.ed=c(3,4),patient.usecondom=0)



# GATHER INFORMATION
gettable <- function(summary1,getanova){
    model4table1 <- rep(NA,9)
    model4table1[1] <- paste("Age (per 10 years)", myformat3(summary1[2,4]), myformat3(summary1[2,6]), myformat3(summary1[2,7]), myformat4(getanova$P[1]),sep=" & ")
    model4table1[2] <- paste("Male", myformat3(summary1[6,4]), myformat3(summary1[6,6]), myformat3(summary1[6,7]), myformat4(getanova$P[2]),sep=" & ")
    model4table1[3] <- paste("District & & & &",myformat4(getanova$P[3]))
    model4table1[4] <- "\\\\hspace{.1in}Namacurra (ref) & 1"
    model4table1[5] <- paste("\\\\hspace{.1in}Alto Molocue", myformat3(summary1[8,4]), myformat3(summary1[8,6]), myformat3(summary1[8,7]),sep=" & ")
    model4table1[6] <- paste("\\\\hspace{.1in}Inhassunge", myformat3(summary1[10,4]), myformat3(summary1[10,6]), myformat3(summary1[10,7]),sep=" & ")
    model4table1[7] <- paste("Education (per level)", myformat3(summary1[4,4]), myformat3(summary1[4,6]), myformat3(summary1[4,7]), myformat4(getanova$P[4]),sep=" & ")
    model4table1[8] <- paste("Use condoms", myformat3(summary1[12,4]), myformat3(summary1[12,6]), myformat3(summary1[12,7]), myformat4(getanova$P[5]),sep=" & ")
    model4table1[9] <- paste("Identify AIDS as common", myformat3(summary1[14,4]), myformat3(summary1[14,6]), myformat3(summary1[14,7]), myformat4(getanova$P[6]),sep=" & ")
    model4table.trans1 <- paste("\\\\hspace{.2in}",model4table1,"\\\\\\\\",sep=" ",collapse=" ")
    return(model4table.trans1)
}
model4table.trans1 <- gettable(trans1.summary.model4,trans1.getanova4)
model4table.trans2 <- gettable(trans2.summary.model4,trans2.getanova4)
model4table.trans5 <- gettable(trans5.summary.model4,trans5.getanova4)
model4table.trans  <- gettable(trans.summary.model4, trans.getanova4 )


# and source of health information
set.seed(50)
pp1 <- datadist(p1) # Necessary in order to get summarization
options(datadist='pp1') # Necessary in order to get summarization - defines ranges for all var.
avoid1.mi <- aregImpute(~patient.avoid1 + patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob,n.impute=10,data=p1)
avoid3.mi <- aregImpute(~patient.avoid3 + patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob,n.impute=10,data=p1)
avoid.mi  <- aregImpute(~as.factor(patient.avoid)  + patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob,n.impute=10,data=p1)

# avoid 1
# avoid1.model9 <- fit.mult.impute(patient.avoid1 ~ rcs(patient.age,3) + patient.sex + patient.dt + patient.ru + rcs(patient.ed,3) + patient.avoid + usecondom, lrm, avoid1.mi, p1)
# p=0.9041 evidence that patient.age is nonlinear with log-odds.
# p=0.1780 evidence that patient.ed is nonlinear with log-odds.    
avoid1.model9 <- fit.mult.impute(patient.avoid1 ~ patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob, lrm, avoid1.mi, p1)
(avoid1.getanova4 <- as.data.frame(anova(avoid1.model9)))
avoid1.model9.stats <- avoid1.model9$stats
avoid1.summary.model9  <- summary(avoid1.model9,patient.age=c(25,35),patient.ed=c(3,4),patient.usecondom=0)

# avoid 3
# avoid3.model9 <- fit.mult.impute(patient.avoid3 ~ rcs(patient.age,3) + patient.sex + patient.dt + patient.ru + rcs(patient.ed,3) + patient.avoid + usecondom, lrm, avoid3.mi, p1)
# p=0.6438 evidence that patient.age is nonlinear with log-odds.
# p=0.0174 evidence that patient.ed is nonlinear with log-odds.    
avoid3.model9 <- fit.mult.impute(patient.avoid3 ~ patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob, lrm, avoid3.mi, p1)
(avoid3.getanova4 <- as.data.frame(anova(avoid3.model9)))
avoid3.model9.stats <- avoid3.model9$stats
avoid3.summary.model9  <- summary(avoid3.model9,patient.age=c(25,35),patient.ed=c(3,4),patient.usecondom=0)
# NO avoid 
# avoid.model9 <- fit.mult.impute(as.factor(patient.avoid) ~ rcs(patient.age,3) + patient.sex + patient.dt + patient.ru + rcs(patient.ed,3) + patient.avoid + usecondom, lrm, avoid.mi, p1)
# p=0.9999 evidence that patient.age is nonlinear with log-odds.
# p=0.9999 evidence that patient.ed is nonlinear with log-odds.    
avoid.model9 <- fit.mult.impute(patient.avoid ~ patient.age + patient.sex + patient.dt + patient.ed + patient.usecondom + patient.aidsprob, lrm, avoid.mi, p1)
(avoid.getanova4 <- as.data.frame(anova(avoid.model9)))
avoid.model9.stats <- avoid.model9$stats
avoid.summary.model9  <- summary(avoid.model9,patient.age=c(25,35),patient.ed=c(3,4),patient.usecondom=0)
model9table.avoid1 <- gettable(avoid1.summary.model9,avoid1.getanova4)
model9table.avoid3 <- gettable(avoid3.summary.model9,avoid3.getanova4)
model9table.avoid  <- gettable(avoid.summary.model9, avoid.getanova4 )




# BY PROVIDERS
# I am hoping in terms of health seeking behavior that those who only go to clinics (or who go first) 
# will have more education and knowledge of HIV transmission and prevention behaviors.
# Those who only go to traditional healers (or go there first) might be from more rural areas, less education, or less knowledge. 

patient.prov.age <- patients.prov$Age
patient.prov.sex <- patients.prov$Gender
patient.prov.dt <- as.factor(patients.prov$district)
patient.prov.ru <- as.factor(patients.prov$rural.urban)
patient.prov.ed <- patients.prov$Highest.level.of.Education
patient.prov.cure <- as.factor(cureaids.rows[cureaids[!is.na(patients.byprov2)]+1])
patient.prov.trans1 <- patient.trans[!is.na(patients.byprov2)]
patient.prov.trans <- as.factor(ifelse(patient.prov.trans1 %in% c("1","2","3","4"),1,0))
patient.prov.trans <- as.numeric(as.character(patient.trans[!is.na(patients.byprov2)]))
p <- data.frame(patients.byprov, patient.prov.age, patient.prov.sex, patient.prov.dt, patient.prov.ru, patient.prov.ed, patient.prov.cure, patient.prov.trans)
pp <- datadist(p) # Necessary in order to get summarization
options(datadist='pp') # Necessary in order to get summarization - defines ranges for all var.
p1.mi <- aregImpute(~ patients.byprov + patient.prov.age + patient.prov.sex + patient.prov.dt + patient.prov.ed + patient.prov.trans ,n.impute=10,data=p)
# model5 <- lrm(patients.byprov ~ patient.age + patient.prov.sex + patient.prov.dt + patient.prov.ru + patient.prov.ed + patient.prov.cure, x=TRUE, y=TRUE)
model5 <- fit.mult.impute(patients.byprov ~ rcs(patient.prov.age,3) + patient.prov.sex + patient.prov.dt + patient.prov.ed + patient.prov.trans, lrm, p1.mi, p)
# p=0.4567 evidence that patient.ed is nonlinear with log-odds.    
# p=0.0163 evidence that patient.age   is nonlinear with log-odds.  
# p=0.9516 evidence that patient.trans   is nonlinear with log-odds.  
(getanova5 <- as.data.frame(anova(model5)))
model5.stats <- model5$stats
summary.model5 <- summary(model5,patient.prov.age=c(25,35),patient.prov.ed=c(3,4))
summary.model5a <- summary(model5,patient.prov.age=c(25,45),patient.prov.ed=c(3,4))

model5table1 <- rep(NA,11)
model5table1[1] <- paste("Age & & & &",myformat4(getanova5$P[1]))
model5table1[2] <- "\\\\hspace{.1in}25 (ref) & 1"
model5table1[3] <- paste("\\\\hspace{.1in}35", myformat3(summary.model5[2,4]), myformat3(summary.model5[2,6]), myformat3(summary.model5[2,7]),sep=" & ")
model5table1[4] <- paste("\\\\hspace{.1in}45", myformat3(summary.model5a[2,4]), myformat3(summary.model5a[2,6]), myformat3(summary.model5a[2,7]),sep=" & ")
model5table1[5] <- paste("Male", myformat3(summary.model5[8,4]), myformat3(summary.model5[8,6]), myformat3(summary.model5[8,7]), myformat4(getanova5$P[3]),sep=" & ")
model5table1[6] <- paste("District & & & &",myformat4(getanova5$P[4]))
model5table1[7] <- "\\\\hspace{.1in}Inhassunge (ref) & 1"
model5table1[8] <- paste("\\\\hspace{.1in}Alto Molocue", myformat3(summary.model5[10,4]), myformat3(summary.model5[10,6]), myformat3(summary.model5[10,7]),sep=" & ")
model5table1[9] <- paste("\\\\hspace{.1in}Namacurra", myformat3(summary.model5[12,4]), myformat3(summary.model5[12,6]), myformat3(summary.model5[12,7]),sep=" & ")
model5table1[10] <- paste("Education (per level)", myformat3(summary.model5[4,4]), myformat3(summary.model5[4,6]), myformat3(summary.model5[4,7]), myformat4(getanova5$P[5]),sep=" & ")
model5table1[11] <- paste("Routes of Transmission (per 1 increase)", myformat3(summary.model5[6,4]), myformat3(summary.model5[6,6]), myformat3(summary.model5[6,7]),myformat4(getanova5$P[6]),sep=" & ")
model5table <- paste(model5table1,"\\\\\\\\",sep=" ",collapse=" ")



patient.learn1 <- ifelse(is.na(learn1),0,learn1)
patient.learn2 <- ifelse(is.na(learn2),0,learn2)
patient.learn3 <- ifelse(is.na(learn3),0,learn3)
patient.learn4 <- ifelse(is.na(learn4),0,learn4)
patient.learn5 <- ifelse(is.na(learn5),0,learn5)
patient.learn6 <- ifelse(is.na(learn6),0,learn6)
patient.learn7 <- ifelse(is.na(learn7),0,learn7)


l1 <- data.frame(patient.trans, patient.avoid, patient.learn1, patient.learn2, patient.learn3, patient.learn4, patient.learn5, patient.learn6, patient.learn7)
ll1 <- datadist(l1) # Necessary in order to get summarization
options(datadist='ll1') # Necessary in order to get summarization - defines ranges for all var.
model6 <- lrm(patient.trans ~ patient.learn1 + patient.learn2 + patient.learn3 + patient.learn4 + patient.learn5 + patient.learn6 + patient.learn7, x=TRUE, y=TRUE, data=l1)
(getanova6 <- as.data.frame(anova(model6)))
model6.stats <- model6$stats
summary.model6 <- summary(model6)
model7 <- lrm(patient.avoid ~ patient.learn1 + patient.learn2 + patient.learn3 + patient.learn4 + patient.learn5 + patient.learn6 + patient.learn7, x=TRUE, y=TRUE, data=l1)
(getanova7 <- as.data.frame(anova(model7)))
model7.stats <- model7$stats
summary.model7 <- summary(model7)

getline <- function(summary,getanova,i,name){
    paste(name, myformat3(summary[2*i,4]), myformat3(summary[2*i,6]), myformat3(summary[2*i,7]),myformat4(getanova$P[i]),sep=" & ")
}
model6table1 <- rep(NA,7)
model6table1[1] <- getline(summary.model6,getanova6,1,"School")
model6table1[2] <- getline(summary.model6,getanova6,2,"Radio")
model6table1[3] <- getline(summary.model6,getanova6,3,"Community Meetings")
model6table1[4] <- getline(summary.model6,getanova6,4,"Community Members")
model6table1[5] <- getline(summary.model6,getanova6,5,"Community Health Workers")
model6table1[6] <- getline(summary.model6,getanova6,6,"Traditional Healers")
model6table1[7] <- getline(summary.model6,getanova6,7,"Clinics")
model6table <- paste(model6table1,"\\\\\\\\",sep=" ",collapse=" ")

model7table1 <- rep(NA,7)
model7table1[1] <- getline(summary.model7,getanova7,1,"School")
model7table1[2] <- getline(summary.model7,getanova7,2,"Radio")
model7table1[3] <- getline(summary.model7,getanova7,3,"Community Meetings")
model7table1[4] <- getline(summary.model7,getanova7,4,"Community Members")
model7table1[5] <- getline(summary.model7,getanova7,5,"Community Health Workers")
model7table1[6] <- getline(summary.model7,getanova7,6,"Traditional Healers")
model7table1[7] <- getline(summary.model7,getanova7,7,"Clinics")
model7table <- paste(model7table1,"\\\\\\\\",sep=" ",collapse=" ")

@

\section{Results}
\subsection{Patients Characteristics}
Table \ref{tab:bysex} summarizes patient characteristics by sex.\\
% Hidden R code chunk --- reading in the data


\begin{center}
\setlength{\tabcolsep}{3pt}
\begin{threeparttable}
\footnotesize
\centering{
\caption{Summary of Patient Characteristics by Sex}
\label{tab:bysex}
\begin{tabular}{p{7.5cm}p{2.5cm}p{2.5cm}p{2.5cm}l}
\hline
   & Female             & Male                  & Combined    & P-value\tnote{a}\\
  & (n=\Sexpr{gender.N[1]})  & (n=\Sexpr{gender.N[2]})  & (n=\Sexpr{sum(gender.N)}) \\
\hline 
\Sexpr{age.m6$line.incmiss}\\
\Sexpr{ed.m6$line}\\
\Sexpr{location.n6$line}\\
\Sexpr{rural.n6$line}\\
\Sexpr{civstat.n6$line}\\
\Sexpr{agemar.m6$line.incmiss}\\
\Sexpr{spouse.m6$line.incmiss}\\
\Sexpr{pspouse.m6$line.incmiss}\\
\Sexpr{child.m6$line}\\
\Sexpr{common.n6$line}\\
\Sexpr{comprob1.n6$line}\\
\Sexpr{comprob2.n6$line}\\
\Sexpr{comprob3.n6$line}\\
How do people get infected with HIV/AIDS?\tnote{c} \\
\Sexpr{trans1.n6$line}\\
\Sexpr{trans2.n6$line}\\
\Sexpr{trans3.n6$line}\\
\Sexpr{trans4.n6$line}\\
\Sexpr{trans5.n6$line}\\
How do you prevent HIV/AIDS infection?\tnote{c}\\
\Sexpr{avoid1.n6$line}\\
\Sexpr{avoid2.n6$line}\\
\Sexpr{avoid3.n6$line}\\
\Sexpr{avoid4.n6$line}\\
\Sexpr{cureaids.n6$line}\\
% \Sexpr{usecondom.n6$line}\\
How do you learn about illness and treatment?\tnote{c}\\
\Sexpr{learn1.n6$line}\\
\Sexpr{learn2.n6$line}\\
\Sexpr{learn6.n6$line}\\
\Sexpr{learn6.n6$line}\\
\Sexpr{learn5.n6$line}\\
\Sexpr{learn6.n6$line}\\
\Sexpr{learn7.n6$line}\\


\hline
\end{tabular}
}
\begin{tablenotes}
\item[a] To compare the distribution of patient characteristics by sex,
 we employ chi-square tests.   Similarly, we use a two-sample rank sum test for continuous
 variables by sex.
\item[b] Continuous variables are reported as medians (interquartile range).
\item[c] Percentages may sum to greater than 100\% because multiple answers may be provided.
\end{tablenotes}
\end{threeparttable}
\end{center}


\vspace{1.25in}
\subsection{Patients who use Traditional Healers}
Table \ref{tab:byuse} summarizes patient characteristics by use of Traditional Healers.
 To identify predictors of traditional healer usage, we use logistic regression.
 We multiply impute age for \Sexpr{sum(is.na(patient.age))} patients missing this data.
  Age is a continuous predictor that
 is nonlinear with log-odds of healer use.  To relax our linearity assumption, we model
 age using a 3 knot restricted cubic spline function (this is why age is summarized in
 reference to another age).  Table \ref{tab:lrm3} summarizes the model effects.  
 Adjusting for sex and education, there are three significant predictors of visiting traditional healers;
 these are, age, district location and response to the existence of AIDS cure question.
 This model has decent predictive ability ($c=\Sexpr{format(round(model3.stats[6],2),nsmall=2)}$).\\

Of the \Sexpr{sum(patients.visitTH)} patients who use traditional healers, we are interested
 in determining whether increased cost of visit was associated with increased patient satisfaction. 
 Median (IQR) cost of visit for unsatisfied patients is \Sexpr{payby0} and for satisfied patients is \Sexpr{payby1}.
 A test for incremental \textbf{cost by satisfaction} gives p=\Sexpr{myformat3(kruskal.test(pay.cash[patients.visitTH==1]~visit.sat[patients.visitTH==1])$p.value)}.\\

Of the \Sexpr{sum(patients.visitTH)} patients who use traditional healers, we are interested
 in determining whether decreased education was associated with increased patient satisfaction. 
 Median (IQR) education level for unsatisfied patients is \Sexpr{edby0} and for satisfied patients is \Sexpr{edby1}.
 A test for incremental \textbf{education by satisfaction} gives p=\Sexpr{myformat4(kruskal.test(patients$Highest.level.of.Education[patients.visitTH==1]~visit.sat[patients.visitTH==1])$p.value)}.



\begin{center}
\footnotesize
\begin{longtable}{p{7.5cm}p{2.5cm}p{2.5cm}p{2.5cm}l}
\caption{Summary of Patient Characteristics and Response by Use of Traditional Healers}\label{tab:byuse}\\
\hline
Do you visit   & No             & Yes                  & Combined    & P-value\tnote{a}\\
 traditional healers? & (n=\Sexpr{patients.visitTH.N[1]})  & (n=\Sexpr{patients.visitTH.N[2]})  & (n=\Sexpr{sum(patients.visitTH.N)}) \\
\hline 
\endfirsthead
\multicolumn{5}{c}{{\tablename} \thetable{} -- Continued} \\[0.5ex]
\hline
Do you visit   & No             & Yes                  & Combined    & P-value\tnote{a}\\
 traditional healers? & (n=\Sexpr{patients.visitTH.N[1]})  & (n=\Sexpr{patients.visitTH.N[2]})  & (n=\Sexpr{sum(patients.visitTH.N)}) \\
\hline  
\endhead
\multicolumn{5}{c}{{Continued on Next Page\ldots}} \\
\endfoot
\hline
\endlastfoot
\Sexpr{age.m3$line.incmiss}\\
\Sexpr{sex.n3$line}\\
\Sexpr{ed.m3$line}\\
\Sexpr{location.n3$line}\\
\Sexpr{rural.n3$line}\\
\Sexpr{civstat.n3$line}\\
\Sexpr{agemar.m3$line.incmiss}\\
\Sexpr{spouse.m3$line.incmiss}\\
\Sexpr{pspouse.m3$line.incmiss}\\
\Sexpr{child.m3$line}\\
\Sexpr{common.n3$line}\\
\Sexpr{comprob1.n3$line}\\
\Sexpr{comprob2.n3$line}\\
\Sexpr{comprob3.n3$line}\\
How do people get infected with HIV/AIDS?\tnote{c} \\
\Sexpr{trans1.n3$line}\\
\Sexpr{trans2.n3$line}\\
\Sexpr{trans3.n3$line}\\
\Sexpr{trans4.n3$line}\\
\Sexpr{trans5.n3$line}\\
How do you prevent HIV/AIDS infection?\tnote{c}\\
\Sexpr{avoid1.n3$line}\\
\Sexpr{avoid2.n3$line}\\
\Sexpr{avoid3.n3$line}\\
\Sexpr{avoid4.n3$line}\\
\Sexpr{cureaids.n3$line}\\
 \Sexpr{usecondom.n3$line}\\
How do you learn about illness and treatment?\tnote{c}\\
\Sexpr{learn1.n3$line}\\
\Sexpr{learn2.n3$line}\\
\Sexpr{learn3.n3$line}\\
\Sexpr{learn4.n3$line}\\
\Sexpr{learn5.n3$line}\\
\Sexpr{learn6.n3$line}\\
\Sexpr{learn7.n3$line}\\
\Sexpr{provmore.n3$line.incmiss}\\
\Sexpr{prov.n3$line.incmiss}\\
Visits a specific TH & - & \Sexpr{visit.specTH.n3} & - \\
Satisfied with last visit & - & \Sexpr{visit.sat.n3} & - \\
How much do you pay ({\it MZN})? & - & \Sexpr{pay.cash.m3} & - \\
\hspace{.1in}Missing/non-cash amount & - & \Sexpr{pay.cash.missing} & - \\
\hline
\end{longtable}
\end{center}
\vspace{-.6in}
\begin{flushleft}
  {\scriptsize $^a$ To compare the distribution of patient characteristics by use of traditional healers,
 we employ chi-square tests.   Similarly, we use a two-sample rank sum test for continuous
 variables by use of traditional healers. \\
     $^b$ Continuous variables are reported as medians (interquartile range).\\
     $^c$ Percentages may sum to greater than 100\% because multiple answers may be provided.}
\end{flushleft}




\begin{center}
%\setlength{\tabcolsep}{4pt}
\begin{threeparttable}
\footnotesize
\caption{Model Effects: Patients who use Traditional Healers}
\label{tab:lrm3}
\centering{
\begin{tabular}{llllll}
\hline
  & Odds Ratio  & Lower 95\% & Upper 95\% & P-value \\
\hline 
\\
\Sexpr{model3table} 
\\
\hline
\end{tabular}
}
\end{threeparttable}
\end{center}
\vspace{2in}

\subsection{Patients who identify routes of transmission}
Table \ref{tab:bytrans} summarizes patient characteristics by those who identify
 zero routes of infection versus those who identify one or more.
 To identify predictors of HIV knowledge, we fit a proportional-odds model with number
 of correct routes of infection as the outcome.
Additionally five separate logistic regressions using the same predictors would
 identify which demographics or health-seeking behavior are predictive of knowledge for
 each of five methods of transmission; however, `Mother to Child' and `Blood' were too rarely identified for modeling. 
 We multiply impute age for \Sexpr{sum(is.na(patient.age))} patients and condom usage for \Sexpr{sum(is.na(patient.usecondom))} patients missing this data.
  Tables \ref{tab:lrmtrans1} and \ref{tab:lrmtrans2} summarize the model effects.  \\
 % These models all had decent predictive ability ($c=\Sexpr{format(round(trans1.model4.stats[6],2),nsmall=2)}$).\\


\begin{landscape}
\begin{center}
\footnotesize
\begin{longtable}{p{7.5cm}p{2.5cm}p{2.5cm}p{2.5cm}p{2.5cm}p{2.5cm}l}
\caption{Summary of Patient Characteristics and Response by Knowledge of HIV Transmission}\label{tab:bytrans}\\
\hline
Identify one of five       & None             & 1 & 2 & 3 or more                  & Combined    & P-value\tnote{a}\\
  routes of transmission? & (n=\Sexpr{patients.trans3.N[1]})  & (n=\Sexpr{patients.trans3.N[2]}) & (n=\Sexpr{patients.trans3.N[3]}) & (n=\Sexpr{patients.trans3.N[4]})  & (n=\Sexpr{sum(patients.trans3.N)}) \\
\hline 
\endfirsthead
\multicolumn{7}{c}{{\tablename} \thetable{} -- Continued} \\[0.5ex]
\hline
Identify one of five       & None             & 1 & 2 & 3 or more                  & Combined    & P-value\tnote{a}\\
  routes of transmission? & (n=\Sexpr{patients.trans3.N[1]})  & (n=\Sexpr{patients.trans3.N[2]}) & (n=\Sexpr{patients.trans3.N[3]}) & (n=\Sexpr{patients.trans3.N[4]})  & (n=\Sexpr{sum(patients.trans3.N)}) \\
\hline  
\endhead
\multicolumn{7}{c}{{Continued on Next Page\ldots}} \\
\endfoot
\hline
\endlastfoot
\Sexpr{age.m4$line.incmiss}\\
\Sexpr{sex.n4$line}\\
\Sexpr{ed.m4$line}\\
\Sexpr{location.n4$line}\\
\Sexpr{rural.n4$line}\\
\Sexpr{civstat.n4$line}\\
\Sexpr{agemar.m4$line.incmiss}\\
\Sexpr{spouse.m4$line.incmiss}\\
\Sexpr{pspouse.m4$line.incmiss}\\
\Sexpr{child.m4$line}\\
\Sexpr{common.n4$line}\\
\Sexpr{comprob1.n4$line}\\
\Sexpr{comprob2.n4$line}\\
\Sexpr{comprob3.n4$line}\\
How do people get infected with HIV/AIDS?\tnote{c} \\
\Sexpr{trans1.n4$line}\\
\Sexpr{trans2.n4$line}\\
\Sexpr{trans3.n4$line}\\
\Sexpr{trans4.n4$line}\\
\Sexpr{trans5.n4$line}\\
How do you prevent HIV/AIDS infection?\tnote{c}\\
\Sexpr{avoid1.n4$line}\\
\Sexpr{avoid2.n4$line}\\
\Sexpr{avoid3.n4$line}\\
\Sexpr{avoid4.n4$line}\\
\Sexpr{cureaids.n4$line}\\
\Sexpr{usecondom.n4$line}\\
How do you learn about illness and treatment?\tnote{c}\\
\Sexpr{learn1.n4$line}\\
\Sexpr{learn2.n4$line}\\
\Sexpr{learn3.n4$line}\\
\Sexpr{learn4.n4$line}\\
\Sexpr{learn5.n4$line}\\
\Sexpr{learn6.n4$line}\\
\Sexpr{learn7.n4$line}\\
\hline
\end{longtable}
\end{center}
\vspace{-.6in}
\begin{flushleft}
  {\scriptsize $^a$ To compare the distribution of patient characteristics by knowledge of HIV transmission (an ordered variable),
 we employ rank sum tests.   Similarly, we use Spearman rank correlation tests for continuous
 variables by knowledge of HIV transmission. \\
     $^b$ Continuous variables are reported as medians (interquartile range).\\
     $^c$ Percentages may sum to greater than 100\% because multiple answers may be provided.}
\end{flushleft}
\end{landscape}


\begin{center}
%\setlength{\tabcolsep}{4pt}
\begin{threeparttable}
\footnotesize
\caption{Proportional-Odds Model Effects: Predictors of Patient Knowledge of HIV Transmission}
\label{tab:lrmtrans1}
\centering{
\begin{tabular}{llllll}
\hline
  & Odds Ratio  & Lower 95\% & Upper 95\% & P-value \\
\hline 
\\
Predictors of suggesting more routes of transmission (proportional-odds):\\
\Sexpr{model4table.trans}
\\
\hline
\end{tabular}
}
\end{threeparttable}
\end{center}

\vspace{0.75in}

\begin{center}
%\setlength{\tabcolsep}{4pt}
\begin{threeparttable}
\footnotesize
\caption{Model Effects: Predictors of Patient Knowledge of HIV Transmission}
\label{tab:lrmtrans2}
\centering{
\begin{tabular}{llllll}
\hline
  & Odds Ratio  & Lower 95\% & Upper 95\% & P-value \\
\hline 
\\
Predictors of suggesting sex as route of transmission:\\
\Sexpr{model4table.trans1} 
\\
\hline 
\\
Predictors of suggesting razors as route of transmission:\\
\Sexpr{model4table.trans2} 
\\
\hline 
\\
Predictors of suggesting needles as route of transmission:\\
\Sexpr{model4table.trans5}  
\\
\hline
\end{tabular}
}
\end{threeparttable}
\end{center}

\clearpage



\subsection{Patients who identify methods of prevention}
Table \ref{tab:byavoid} summarizes patient characteristics by those who identify
 zero, one, two or more methods of prevention.
 To identify predictors of knowledge of HIV prevention, we fit a proportional-odds model with number
 of correct methods of prevention as the outcome.
Additionally four separate logistic regressions using the same predictors would
 identify which demographics or health-seeking behavior are predictive of knowledge for
 each of four methods of prevention; however, `Do not share razors' and `Abstinence' were too rarely identified for modeling. 
 We multiply impute age for \Sexpr{sum(is.na(patient.age))} patients and condom usage for \Sexpr{sum(is.na(patient.usecondom))} patients missing this data.
  Tables \ref{tab:lrmavoid1} and \ref{tab:lrmavoid2} summarize the model effects.  \\


\begin{landscape}
\begin{center}
\footnotesize
\begin{longtable}{p{7.5cm}p{2.5cm}p{2.5cm}p{2.5cm}p{2.5cm}l}
\caption{Summary of Patient Characteristics and Response by Knowledge of HIV Prevention}\label{tab:byavoid}\\
\hline
Identify one of four       & None             & 1 & 2 or more                  & Combined    & P-value\tnote{a}\\
  methods of prevention? & (n=\Sexpr{patients.avoid3.N[1]})  & (n=\Sexpr{patients.avoid3.N[2]}) & (n=\Sexpr{patients.avoid3.N[3]}) & (n=\Sexpr{sum(patients.avoid3.N)}) \\
\hline 
\endfirsthead
\multicolumn{6}{c}{{\tablename} \thetable{} -- Continued} \\[0.5ex]
\hline
Identify one of four       & None             & 1 & 2 or more                  & Combined    & P-value\tnote{a}\\
  methods of prevention? & (n=\Sexpr{patients.avoid3.N[1]})  & (n=\Sexpr{patients.avoid3.N[2]}) & (n=\Sexpr{patients.avoid3.N[3]}) & (n=\Sexpr{sum(patients.avoid3.N)}) \\
\hline  
\endhead
\multicolumn{6}{c}{{Continued on Next Page\ldots}} \\
\endfoot
\hline
\endlastfoot
\Sexpr{age.m7$line.incmiss}\\
\Sexpr{sex.n7$line}\\
\Sexpr{ed.m7$line}\\
\Sexpr{location.n7$line}\\
\Sexpr{rural.n7$line}\\
\Sexpr{civstat.n7$line}\\
\Sexpr{agemar.m7$line.incmiss}\\
\Sexpr{spouse.m7$line.incmiss}\\
\Sexpr{pspouse.m7$line.incmiss}\\
\Sexpr{child.m7$line}\\
\Sexpr{common.n7$line}\\
\Sexpr{comprob1.n7$line}\\
\Sexpr{comprob2.n7$line}\\
\Sexpr{comprob3.n7$line}\\
How do people get infected with HIV/AIDS?\tnote{c} \\
\Sexpr{trans1.n7$line}\\
\Sexpr{trans2.n7$line}\\
\Sexpr{trans3.n7$line}\\
\Sexpr{trans4.n7$line}\\
\Sexpr{trans5.n7$line}\\
How do you prevent HIV/AIDS infection?\tnote{c}\\
\Sexpr{avoid1.n7$line}\\
\Sexpr{avoid2.n7$line}\\
\Sexpr{avoid3.n7$line}\\
\Sexpr{avoid4.n7$line}\\
\Sexpr{cureaids.n7$line}\\
\Sexpr{usecondom.n7$line}\\
How do you learn about illness and treatment?\tnote{c}\\
\Sexpr{learn1.n7$line}\\
\Sexpr{learn2.n7$line}\\
\Sexpr{learn3.n7$line}\\
\Sexpr{learn4.n7$line}\\
\Sexpr{learn5.n7$line}\\
\Sexpr{learn6.n7$line}\\
\Sexpr{learn7.n7$line}\\
\hline
\end{longtable}
\end{center}
\vspace{-.6in}
\begin{flushleft}
  {\scriptsize $^a$ To compare the distribution of patient characteristics by knowledge of HIV prevention (an ordered variable),
 we employ rank sum tests.   Similarly, we use Spearman rank correlation tests for continuous
 variables by knowledge of HIV transmission. \\
     $^b$ Continuous variables are reported as medians (interquartile range).\\
     $^c$ Percentages may sum to greater than 100\% because multiple answers may be provided.}
\end{flushleft}
\end{landscape}


\begin{center}
%\setlength{\tabcolsep}{4pt}
\begin{threeparttable}
\footnotesize
\caption{Proportional-Odds Model Effects: Predictors of Patient Knowledge of HIV Prevention}
\label{tab:lrmavoid1}
\centering{
\begin{tabular}{llllll}
\hline
  & Odds Ratio  & Lower 95\% & Upper 95\% & P-value \\
\hline 
\\
Predictors of suggesting more methods of prevention (proportional-odds):\\
\Sexpr{model9table.avoid}
\\
\hline
\end{tabular}
}
\end{threeparttable}
\end{center}

\vspace{0.75in}

\begin{center}
%\setlength{\tabcolsep}{4pt}
\begin{threeparttable}
\footnotesize
\caption{Model Effects: Predictors of Patient Knowledge of HIV Prevention}
\label{tab:lrmavoid2}
\centering{
\begin{tabular}{llllll}
\hline
  & Odds Ratio  & Lower 95\% & Upper 95\% & P-value \\
\hline 
\\
Predictors of suggesting condom use for prevention:\\
\Sexpr{model9table.avoid1} 
\\
\hline 
\\
Predictors of suggesting monogamy for prevention:\\
\Sexpr{model9table.avoid3} 
\\
\hline 
\end{tabular}
}
\end{threeparttable}
\end{center}


\clearpage
\subsection{Knowledge of HIV Transmission by How Patients Learn About Illness and Treatment}
 To identify which sources of learning predict increased understanding of HIV transmission, 
 we fit a proportional-odds model with number of correct routes of infection as the outcome.
  Table \ref{tab:lrm6} summarizes the model effects.  \\

\begin{center}
%\setlength{\tabcolsep}{4pt}
\begin{threeparttable}
\footnotesize
\caption{Model Effects: Knowledge of HIV Transmission by Source of Knowledge}
\label{tab:lrm6}
\centering{
\begin{tabular}{llllll}
\hline
  & Odds Ratio  & Lower 95\% & Upper 95\% & P-value \\
\hline 
\\
\Sexpr{model6table} 
\\
\hline
\end{tabular}
}
\end{threeparttable}
\end{center}
\vspace{0.25in}

\subsection{Knowledge of HIV Prevention by How Patients Learn About Illness and Treatment}
 To identify which sources of learning predict increased understanding of HIV prevention, 
 we fit a proportional-odds model with number of correct methods of preventing HIV/AIDS as the outcome.
  Table \ref{tab:lrm7} summarizes the model effects.  \\

\begin{center}
%\setlength{\tabcolsep}{4pt}
\begin{threeparttable}
\footnotesize
\caption{Model Effects: Knowledge of HIV Prevention by Source of Knowledge}
\label{tab:lrm7}
\centering{
\begin{tabular}{llllll}
\hline
  & Odds Ratio  & Lower 95\% & Upper 95\% & P-value \\
\hline 
\\
\Sexpr{model7table} 
\\
\hline
\end{tabular}
}
\end{threeparttable}
\end{center}
\vspace{0.25in}


%\vspace{1.25in}
\subsection{Patients and Provider Usage}
Table \ref{tab:byprov} summarizes patient characteristics by provider usage. Forty-eight patients
 who said they would go to both providers at once were dropped along with 26 who refused to answer
 or responded with a non-provider.


\begin{center}
\setlength{\tabcolsep}{3pt}
\begin{threeparttable}
\footnotesize
\centering{
\caption{Summary of Patient Characteristics and Provider Usage}
\label{tab:byprov}
\begin{tabular}{p{7.5cm}p{2.5cm}p{2.5cm}p{2.5cm}l}
\hline
How do you decide when to    & Hospital only/first         & TH only/first                  & Combined    & P-value\tnote{a}\\
  go to TH or clinic? & (n=\Sexpr{patients.byprov.N[1]})  & (n=\Sexpr{patients.byprov.N[2]})  & (n=\Sexpr{sum(patients.byprov.N)}) \\
\hline 
\Sexpr{age.m5$line.incmiss}\\
\Sexpr{sex.n5$line}\\
\Sexpr{ed.m5$line}\\
\Sexpr{location.n5$line}\\
\Sexpr{rural.n5$line}\\
\Sexpr{civstat.n5$line}\\
\Sexpr{agemar.m5$line.incmiss}\\
\Sexpr{spouse.m5$line.incmiss}\\
\Sexpr{pspouse.m5$line.incmiss}\\
\Sexpr{child.m5$line}\\
\Sexpr{common.n5$line}\\
\Sexpr{comprob1.n5$line}\\
\Sexpr{comprob2.n5$line}\\
\Sexpr{comprob3.n5$line}\\
How do people get infected with HIV/AIDS?\tnote{c} \\
\Sexpr{trans1.n5$line}\\
\Sexpr{trans2.n5$line}\\
\Sexpr{trans3.n5$line}\\
\Sexpr{trans4.n5$line}\\
\Sexpr{trans5.n5$line}\\
How do you prevent HIV/AIDS infection?\tnote{c}\\
\Sexpr{avoid1.n5$line}\\
\Sexpr{avoid2.n5$line}\\
\Sexpr{avoid3.n5$line}\\
\Sexpr{avoid4.n5$line}\\
\Sexpr{cureaids.n5$line}\\
% \Sexpr{usecondom.n5$line}\\
How do you learn about illness and treatment?\tnote{c}\\
\Sexpr{learn1.n5$line}\\
\Sexpr{learn2.n5$line}\\
\Sexpr{learn3.n5$line}\\
\Sexpr{learn4.n5$line}\\
\Sexpr{learn5.n5$line}\\
\Sexpr{learn6.n5$line}\\
\Sexpr{learn7.n5$line}\\

\hline
\end{tabular}
}
\begin{tablenotes}
\item[a] To compare the distribution of patient characteristics by provider usage,
 we employ chi-square tests.   Similarly, we use a two-sample rank sum test for continuous
 variables by provider usage.
\item[b] Continuous variables are reported as medians (interquartile range).
\item[c] Percentages may sum to greater than 100\% because multiple answers may be provided.
\end{tablenotes}
\end{threeparttable}
\end{center}

\begin{center}
%\setlength{\tabcolsep}{4pt}
\begin{threeparttable}
\footnotesize
\caption{Model Effects: Patients and Provider Usage}
This is the odds of visiting TH only/first versus hospital only/first.
\label{tab:lrm5}
\centering{
\begin{tabular}{llllll}
\hline
  & Odds Ratio  & Lower 95\% & Upper 95\% & P-value \\
\hline 
\\
\Sexpr{model5table} 
\\
\hline
\end{tabular}
}
\end{threeparttable}
\end{center}


\end{document}