Dataset
This study used four datasets: the 2018 NHIS, the 2014 NSDUH, and the 2013 BRFSS in the United States, as well as the 2016 Comprehensive Survey of Living Conditions (CSLC) in Japan. The NHIS, NSDUH, and BRFSS are the major surveys for mental health indicators in the United States mainly sponsored by the National Center for Health Statistics, the Substance Abuse and Mental Health Services Administration, and the Centers for Disease Control and Prevention, respectively. The 2018 NHIS, 2014 NSDUH, 2013 BRFSS, and 2016 CSLC were used because they are the latest available K6 datasets from each survey: the BRFSS has not included the K6 after the 2013 BRFSS and the CSLC had not included the K6 in 2017 and 2018.
NHIS is a continuous nationally representative sample survey that collects data on a broad range of health measures using personal house interviews [
15]. NSDUH is the primary source of the use of illegal drugs by the US population and also includes indicators of mental health problems [
16]. BRFSS is a state-based system of health surveys that collect information on health-risk behaviors, clinical preventive practices, and health care access [
17]. CSLC is a repeated national cross-sectional survey that evaluates the basic living conditions of residents in Japan [
18]. NHIS, NSDUH, and BRFSS datasets are accessible to researchers worldwide through their official repositories. The CSLC dataset is accessible to researchers upon approval by the Ministry of Health, Labour and Welfare of Japan through specific application procedures.
The NHIS is a multistage probability sample survey of the civilian, noninstitutionalized US population aged ≥18 years. The NHIS sample included 25,417 respondents, and the total household response rate was 64.2%. The NSDUH uses a national probability sample of the US civilian, noninstitutionalized population aged ≥12 years (18 years or older for this analysis). The NSDUH sample included 55,272 respondents, with a response rate of 71.2%.
The BRFSS is a population-based state surveillance system that uses random-digit-dialed telephone surveys of noninstitutionalized US citizens aged ≥18 years. The BRFSS questionnaire includes optional modules on specific topics. In 2013, four states: Minnesota, Nevada, Tennessee, and Washington used an optional module of Mental Illness and Stigma that included the K6. The total 2013 K6 sample comprised 33,211 respondents, including 12,781, 4564, 5611, and 10,255 respondents for Minnesota, Nevada, Tennessee, and Washington, with response rates of 54.3, 43.7, 45.9, and 31.1%, respectively.
The survey unit of CSLC includes households and household members; they are randomly selected by stratified random sampling method of census enumeration districts in Japan, covering approximately 300,000 households. The response rate of the survey was 77.5%. This study used CSLC data from a total of 220,294 participants (≥18 years old). Thus, all the participants for the four datasets were adults aged ≥18 years. The socio-demographic characteristics of the participants of the four surveys have been reported in detail elsewhere [
19‐
22].
Measures
The standard K6 includes six items related to the degree with which participants have felt (1) nervous, (2) hopeless, (3) restless or fidgety, (4) so depressed that nothing could cheer them up, (5) that everything was an effort, and (6) worthless over the previous 30 days [
1]. Each item was self-rated on a 5-point scale ranging from 0 = “none of the time” to 4 = “all of the time,” yielding a total item score range of 0–24. Respondents with a total score of 13 or greater are classified as having past month serious psychological distress (SPD). Although all surveys use the K6 to assess past month psychological distress and the same cutoff score for SPD, variations exist in the wording of question items among the surveys. For example, the fourth K6 item used in the NHIS survey is worded as “
feel so sad,” while the one in the NSDUH is “
feel so sad or depressed” and the one in the BRFSS is “
feel so depressed.” The CSLC uses the Japanese version of the K6, which is developed in accordance with the World Health Organization (WHO) translation guidelines and shown to have screening performances equivalent to that of the original English version [
23].
Analysis
First, we calculated the descriptive statistics of the K6 score distributions for the NHIS, NSDUH, BRFSS, and CSLC. As our hypothesis was not based on equality of distributions, but on similarity of distribution patterns regardless of regardless of different sample populations and methodologies, we did not estimate significant differences based on the null hypothesis that the mean values and PSD percentages were equal.
Next, we compared the goodness of fit between the three models: exponential, power law, and quadratic function models. The Corrected Akaike’s Information Criterion (AICc), Schwarz’s Bayesian Information Criterion (BIC), and Root Mean Square Error (RMSE) were calculated to compare the fits. With regard to AICc and BIC, models with smaller values are considered to be a better fit. A fitting curve was calculated using the method of least squares.
Moreover, we graphically analyzed the patterns of the K6 total score distributions from the four surveys using normal and logarithmic scales. Although histograms are generally used for ordinal variables, we used line charts representing the K6 total score distributions because we cannot plot multiple histograms together on a single graph. Thus, we used line charts to show the common mathematical pattern across the four surveys. All statistical procedures were performed using JMP Version 15.0.0 for Windows (SAS Institute, Inc., Cary, NC, US). Because the purpose of the present research was to investigate a mathematical pattern of sampling distribution, all analyses were based on raw data.