This study was a cross-sectional, population-based study and the primary source of data was from the Korean National Health and Nutrition Examination Survey (KNHANES) from 2007–2012. KNHANES 2007–2012 study years were chosen because during this period the questionnaire system of data collection was used for cancer screening and information about the last time participants had screening was also included in the questionnaire. The target population of KNHANES comprises noninstitutionalized Korean citizens residing in Korea and the sampling plan followed by a multistage clustered probability design. Independent variables included were urban and rural areas classified according to the Korean Local Government Act. Korea consists of metropolitan cities, provinces and nonmetropolitan cities divided into districts, and provinces divided into cities and counties. The cities and districts include multiple “dong” and counties including multiple “eup” and “myeon” hence in this study we classified dong as urban areas and eup and myeon as rural areas respectively as independent variables.
This study included men and women of ages ≥ 40 years, women of ages ≥ 40 years and women of ages ≥ 30 for gastric, breast and cervical cancers respectively with no previous cancer diagnosis in all three cancers living in rural or urban areas according to the target population of NCSP (Supplementary Table 1). Amongst a total of 4,594 participants in 2007 gastric cancer having 1,984 participants, 1,126 for breast cancer and 1,490 for cervical cancer were involved in the final analysis after exclusion of participants with previous cancer history, lower age group, missing values in cancer screening and inspection methods. The same method was followed for the 2008, 2009 and 2010 years.
In this study, we considered dependent variables based on previous studies on cancer screening rates in rural and urban areas [9–11]. Socio-demographic variables involved in this study were age, education, income, and medical insurance. Medical insurance includes the National Health Insurance (NHI) which is compulsory to all Koreans residents and the Medical Aid Program put in place by the government to secure the minimum living standards to low income-households [12].
The general characteristics of study participants and cancer screening rate were calculated using summary statistics and weighted frequencies and percentages of screening rates, with 95% confidence intervals, were calculated for the three different cancers in rural and urban areas using chi-squared tests. We adjusted for age after dividing the number of ever cancer screened participants by the total number of screened and unscreened population in each age category as a percentage and multiplied by age-specific weights using the Korea 2000 mid-year as standard population within each age category. The weighted rates were later summed up to give the age-adjusted screening rates in urban and rural areas.
Trends in cancer screening rates were calculated using the Annual Percentage Change (APC) as an estimator in the Korean population from 2007 to 2012 in urban and rural areas. Changes in annual screening rates with 95% confident intervals in rural and urban areas were calculated after adjusting for age using the Korea 2000 mid-year as standard population.
Annual percent change as an estimator
Denote the observed screening rate at time ti as ri with the associated random variable Ri, and denote the corresponding expected rate as yi= E(Ri/ti), where n observed ordered time points are a = t1 < t2……<tn = b. Assume that ti represents years and is equally distributed over [a,b] with tj+1 = tj + 1. Also assume that log (γi) is linear over the entire time interval [a,b], that is log (γi) = βo + βti for I = 1……n. Then the annual rate of change is γj+1/γj = exp(β). (γ-j -γ- j)/γ-j =exp(β)-1 and the conventional APC is APC = [exp(β)-1] [13].
Socio-economic characteristics including age range, income status, education levels and residential areas on recommended and ever screening rates from 2007–2009 and 2010–2012 were analyzed using the multivariable logistic regression. This was done to verify which socio-economic factors are more liable to influence screening in urban and rural areas. A p-value of < 0.05 was considered significant for all analysis. All analyses were conducted using STATA software version 12.0.