Predicting Fracture Risk Factors in SAS

## BACKGROUND

Osteoporosis is a common chronic disease of public health concern that results in fragility fractures affecting one in three women and one in five men over age 50 years. (IOF, 2020) This imposes a burden on the individuals and their caretakers result in morbidity and reduced quality of life.. Individuals who sustained osteoporotic fractures endure back pain, loss of height, loss of mobility and many other adverse effects on the quality of their lives. Over 1.5 million fractures per year occurs due to osteoporosis with 250,000 hip fracture, 500,000 hospitalizations, 800,000 emergency room visits, 2,600,000 physician visits, and 180,000 nursing home placements. This cost the Unites states $20 billion in 2015 (Rodan, 2000). Hip fractures are a leading cause of morbidity and loss of independence in the older population and a 1-year mortality of 20% (Johnell,2006). Osteoporotic fractures had resulted in an estimated loss of 5.8 million disability adjusted life years (DALYs) yearly (Johnell,2006).

These staggering cost on the health care system and the excruciating burden on the patient and the healthcare system need to be examined by public health professionals and medical personnel. Osteoporosis can be treated, and fractures prevented.

There are multiple factors that may explain the relationship between race/ethnicity and fracture risks. Race/ethnicity is a result of genetics which dictates the bone density, skeletal size, peak bone mass and body composition.  Socio economic factors are as a result of disparities in social position in the society. (William, 2012) Those disparities influence the amount of sunlight exposure (related to education, income, neighborhood, crime rate, availability of parks, bicycle trails). The socioeconomic status also influences availability of primary prevention resources, smoking habits, alcohol consumption, comorbidities and proper diets. All these factors are interrelated as shown in the figure below:

![Conceptualframework for mechanisms through which race/ethnicity may affect osteoporotic fracture risks](/img/posts/sas/2.png)


## AIM

-	To determine the association between race and osteoporosis fracture
-	To identify factors associated with osteoporosis fractures.

## METHODS

Data source is the National Health and Nutrition Examination Survey (NHANES) 2017- 2018, a nationally representative, cross-sectional survey designed to assess the health and nutritional status of adults and children across the United States. NHANES is a major program of the National Center for Health Statistics (NCHS) and is responsible for producing vital and health statistics for the Nation. (NCHS, 2017). The survey examines a national representative sample of individuals located in counties across the country.
The survey has examined a nationally representative sample of about 5,000 persons each year since
early 1960. These persons are in counties across the country, 15 of which are visited each year.
The NHANES interview includes demographic, socioeconomic, dietary, and health-related
questions. The examination component consists of medical, dental, and physiological measurements, as well as laboratory tests administered by highly trained medical personnel. This study uses data from the National Health and Nutrition Examination Survey, 2017-2018
(NHANES 2017-2018), collected between January 2017 and December 2018.

## Predictors
Fractures is the response variable.
The predictors include baseline characteristics, clinical risk factors and laboratory findings.
Gender, Age, Race, Marital status, Education, Calcium, Diabetes, Arthritis, Stroke, BMI,Alcohol intake, osteoporosis, prednisolone,family history of osteoporosis, smoking, smokingfrequency, T-Score of femur neck, femur neck bone mineral density, vigorous and moderate physical activity

## Data Preprocessing and Statistical Methods:
Analysis was conducted using RStudio Version 1.3.1073 and SAS studio. Data cleaning and preprocessing was done on RStudio while the data analysis was don on SASstudio.

The variables needed from NHANES 2017-2018 was selected, merged, and formatted
(Demographics, diet, smoking, alcohol, diabetes, fractures, osteoporosis, pregnancy, BMI, femoral bone mineral density, spine bone mineral density, calcium, and vitamin d levels, physical activity). This resulted in a total of 9253 individuals with 33 predictors.

The analysis was limited to individuals aged ≥50 years who were both interviewed and examined (3069 participants).
Subjects with missing data on bone mineral density (BMD) and fractures was excluded which makes the sample size to be 2286 with 28 predictors.
Variables with missing values of more than 40%: vit D- 52%, alcohol- 42%, smoking frequency- 54%, T-score of spine-49%, Spine Bone Mineral Density -49%, hormone replacement therapy-
55%. Poverty-income-ratio with missing value of 13% was also removed. One observation missing more than 2 features was removed as well. These leaves us with a data set of 2284 observations and 20 predictors. Each of the features had missing value of less than 5%. The sum of the missing values was 423 (18.5%) with a mean of 0.00925%


## Statistical Methods
Univariate analysis was conducted to describe the sample and estimate the prevalence of the racial groups and the prevalence of osteoporotic fractures. Descriptive statistics was calculated as percentages to describe the characteristics of the sample. Bivariate analysis was conducted thereafter to evaluate the relationship between race/ethnicity and osteoporotic fractures, tested using χ2 tests for proportions as appropriate.
Significant variables from the bivariate analysis were included in the stratified analysis (race, osteoporosis, smoking, alcohol intake, gender, T-score of femur neck, family history and age). Single factor stratification analysis to assess to for effect modification on the multiplicative and additive scale and for confounding using the Breslow- Day test of homogeneity, the stratum specific odds ratio and adjusted odds ratio ratios. Thereafter, multiple factors stratified analysis by smoking and gender was conducted to assess for effect modification 2 factors and for joint confounding.
I tested differences by gender, smoking and age in the association between race/ethnicity and osteoporotic fractures for significance by including first-order interaction terms (e.g., race*gender race*smoking race*age) in linear regression models. The interactions testing gender, smoking or age differences in the relationship between race/ethnicity and osteoporotic fractures were not significant at P < 0.05, so I opted not to include the interaction terms in the final models. Linear regression analysis was used when appropriate to assess the potential effect of race/ethnicity and osteoporotic fractures, after adjusting for race/ethnicity, gender, age and smoking, family history of osteoporosis, and osteoporosis diagnosis. Odds ratio is estimated based on logit- binomial distribution in the multivariable analysis. Descriptive and regression analyses will be conducted using SAS while all other statistical analyses will be performed with R software, and p < 0.05 is considered statistically significant. 



```{sas}

FILENAME REFFILE '/home/u42976927/FinalProject.csv';

PROC IMPORT DATAFILE=REFFILE
	DBMS=CSV
	OUT=WORK.dat
	replace;
	GETNAMES=YES;
	
RUN;


data dat (drop=VAR1 femurneckBMD) ;
set WORK.dat;
run;

/* column list extracted */
proc contents data=dat out=column_nm varnum;
run;

proc print data= dat;
run;

-------------------------------------------------------------
/* Format the variables */

proc format;

value frac
    1 = ' yes'
    0 = 'no'
;

value $race
'Non-Hispanic White' ='    White, NH'
'Non-Hispanic Blacks'='  Black, NH'
'Non-Hispanic Asian'= 'Asian, NH'
'Hispanic'=           '   Hispanic'
'Other Race'=         ' Other Race'
;

value $gender
    'Male' = ' Male'
    'Female' = 'Female'
;
value $Marital
    'Married' = ' Married'
    'Never Married' = '   Never Married'
	'Previously Married'='  Previously Married'
	'NA' = 'Missing'
;
value $calc
    'high ca' = 'Missing'
    'normal ca' = '  normal ca'
	'low ca'='   low ca'
	'NA' = 'Missing'
;

value $diab
    'Yes' = '  Yes'
    'No' = ' No'
	'NA' = 'Missing'
;
	
value $arth
    'Yes' = '  Yes'
    'No' = ' No'
	'NA' = 'Missing'
;

value $strok
    'Yes' = '  Yes'
    'No' = ' No'
	'NA' = 'Missing'
;

value $bmi
    'underweight' =  '    underweight'
    'normalweight' = '   normalweight'
    'overweight' = '  overweight'
    'obese' =  ' obese'
	'NA' = 'Missing'
;

value $alq
    'Yes' = '  Yes'
    'No' = ' No'
	'NA' = 'Missing'
;

value $osteo
    'Yes' = '  Yes'
    'No' = ' No'
	'NA' = 'Missing'
;

value $pred
    'Yes' = '  Yes'
    'No' = ' No'
	'NA' = 'Missing'
;

value $fmhx
    'Yes' = '  Yes'
    'No' = ' No'
	'NA' = 'Missing'
;

value $smk
    'Yes' = ' Yes'
    'No' = 'No'
;

value $edu
    'noHS' =  '   noHS'
    'some/completedHS' = '  some/completedHS'
    'highereducation' = ' highereducation'
	'NA' = 'Missing'
;

value $Tscorefemurneck
    'low bone mass' =  ' low bone mass'
    'normal' = 'normal'
    'osteo' = '  osteo'
	
;
value $vig
    'Vigorous' = '  Yes'
    'NoVigorous' = ' No'
	
;	
value $mod
    'Moderate' = '  Yes'
    'NoModerate' = ' No'
	
;	
run;


*Univariate analysis Table 1*;
proc freq data=dat order= formatted;
format frac frac. calc $calc. ;
tables calc*frac;
run;

*Bivariate Table 2*;

proc freq data=dat order=formatted;
format  frac frac. Gender $gender. race $race. Marital $Marital. calc $calc. 
	diab $diab. arth $arth. strok $strok. bmi $bmi. alq $alq. osteo $osteo. pred $pred. 
	fmhx $fmhx. smk $smk. edu $edu. vig $vig. mod $mod.;
tables mod*frac /cmh ;
run;

proc genmod data=dat order=formatted ;
format  frac frac. Gender $gender. race $race. Marital $Marital. calc $calc. 
	diab $diab. arth $arth. strok $strok. bmi $bmi. alq $alq. osteo $osteo. pred $pred. 
	fmhx $fmhx. smk $smk. edu $edu.vig $vig. mod $mod. Tscorefemurneck $Tscorefemurneck.;
class race (ref='White, NH') osteo(ref='No') smk(ref='No') calc(ref='normal ca') alq(ref='No') 
Gender(ref='Female')strok(ref='No') pred(ref='No') bmi(ref='normalweight') arth (ref= 'No')
Marital (ref='Married') edu(ref='highereducation') diab (ref='No')Tscorefemurneck(ref='normal') 
fmhx(ref='No') age (ref='50-59')/param=ref; 
model frac= bmi /  dist=bin;

estimate "RR White" bmi    1 0 0 0   /exp;
estimate "RR Hispanic" bmi 0 1  0 0 /exp;
estimate "RR Black" bmi    0 0 1 0 /exp;
estimate "RR Black" bmi    0 0 0 1 /exp;


proc freq data=dat order=formatted;
format  frac frac. Gender $gender. race $race. Marital $Marital. calc $calc. 
	diab $diab. arth $arth. strok $strok. bmi $bmi. alq $alq. osteo $osteo. pred $pred. 
	fmhx $fmhx. smk $smk. edu $edu. vig $vig. mod $mod.;
tables Gender age race Marital calc diab arth strok
	bmi alq osteo pred fmhx smk edu Tscorefemurneck vig mod ;
run;

/*General association chi-square test*/;
proc freq data=dat order=formatted;
format  frac frac. Gender $gender. race $race. Marital $Marital. calc $calc. 
	diab $diab. arth $arth. strok $strok. bmi $bmi. alq $alq. osteo $osteo. pred $pred. 
	fmhx $fmhx. smk $smk. edu $edu.vig $vig. mod $mod.;
tables mod*frac/chisq; 
run; 

*Test of trend*;
proc freq data=dat order=formatted;
format  frac frac. ;
tables Tscorefemurneck*frac/trend norow nocol nopercent scores=table; 
run;

*Bivariate, crude rr*;
*using black as reference*;
proc freq data=dat order= formatted;
format frac frac. race $race. ; 
tables race*frac/ cmh ;
where race in ('Non-Hispanic White','Non-Hispanic Blacks');
run;

*using whites as reference*;
proc freq data=dat order= formatted;
format frac frac.  ; 
tables race*frac/ cmh;
where race in ('Non-Hispanic White','Non-Hispanic Blacks');
run;


*/Table 3 stratified analysis with breslow day test/*;

proc contents data=dat;
run;

data datt;
set dat;
if race = 'Non-Hispanic Blacks' then race = 'poc';
if race = 'Hispanic' then race = 'poc';
if race = 'Non-Hispanic White' then race = 'white';
run;

proc freq data=datt;
tables race*frac;
run;

#*bivariate poc vs whites*;#
proc freq data=datt order= formatted;
format frac frac.  ; 
tables race*frac/ cmh;
where race in ('poc','white');
run;


proc freq data= datt order=formatted;
format  frac frac. Gender $gender.  Marital $Marital. calc $calc. 
	diab $diab. arth $arth. strok $strok. bmi $bmi. alq $alq. osteo $osteo. pred $pred. 
	fmhx $fmhx. smk $smk. edu $edu.vig $vig. mod $mod. Tscorefemurneck $Tscorefemurneck.;
table fmhx*race*frac/OR cmh ;
where race in ('poc','white');
run;



*Final Table 4*;

proc genmod data=datt order=formatted ;
format  frac frac. Gender $gender.  Marital $Marital. calc $calc. 
	diab $diab. arth $arth. strok $strok. bmi $bmi. alq $alq. osteo $osteo. pred $pred. 
	fmhx $fmhx. smk $smk. edu $edu.vig $vig. mod $mod. Tscorefemurneck $Tscorefemurneck.;
class race (ref='white') osteo(ref='No') smk(ref='No') calc(ref='normal ca') alq(ref='No') 
Gender(ref='Female') Tscorefemurneck(ref='normal') fmhx(ref='No') age (ref='50-59')/param=ref; 
model frac= Race Gender age smk fmhx alq Tscorefemurneck / link=logit dist=bin ;
where race in ('poc','white');
estimate "POC" race 1  /exp;
estimate "male" Gender 1  /exp;
estimate "60-69" age 1 0  /exp;
estimate ">70" age  0 1 /exp;
estimate "smoking" smk 1 /exp;

estimate "Fam Hx" fmhx 1 0 /exp;
estimate "Alc" alq 1 0 /exp;
estimate "osteo" Tscorefemurneck 1 0 /exp;
estimate "lbm" Tscorefemurneck 0 1/exp;
run;