Tobacco and alcohol use, fruit and vegetable consumption, physical activity habits, and participation in routine checkups and cancer screenings among adults aged 40 and older with and without chronic diseases Authors: Elif Nur Yildirim-Ozturk, Mehmet Uyar Published: 2026 ABSTRACT The study aimed to investigate tobacco and alcohol use, fruit and vegetable consumption, physical activity habits, and participation in routine checkups and cancer screenings in adults aged 40 years and older with and without a diagnosis of chronic disease. It also aimed to compare the frequentist and Bayesian statistical methods used for comparisons and to evaluate the practical significance of these associations using effect sizes. The study was cross-sectional. The data used in the study came from the Turkey Health Survey-2022 and 12,555 people were included in this study. A total of 29 variables were used, and chronic diseases was taken as a grouping variable. Both frequentist and Bayesian methods, including bivariate analyses and multivariable logistic regression models, were utilized. In the study group, 52% were female, 42.8% were overweight, and 46.8% had good general health. Overall, the findings indicate that people with chronic diseases tend to use less tobacco and alcohol, do less physical activity, and have more routine checkups and cancer screenings. Fruit and vegetable/salad consumption was similar between the groups. While frequentist and Bayesian bivariate tests yielded similar results, multivariable logistic regression highlighted critical divergences, with the Bayesian approach acting as a more conservative filter against weak predictors. Furthermore, evaluating the absolute magnitude of these associations revealed strictly small to moderate effect sizes. This study suggests that while behavioral shifts following a chronic disease diagnosis are statistically detectable, the real-world changes might be modest. Public health interventions could benefit from offering targeted programs tailored to patients’ limitations rather than generic advice. Methodologically, future health research and policy-making may benefit from placing more emphasis on effect sizes alongside frequentist or Bayesian approaches. 1. Introduction Chronic diseases – noncommunicable diseases characterized by long duration and slow progression – are the leading cause of the global burden of disease. This group of diseases is estimated to cause 41 million deaths annually, accounting for 74% of all deaths worldwide. This group includes many diseases that are not transmitted from person to person, such as cardiovascular diseases, cancers, chronic obstructive pulmonary disease (COPD), diabetes, and mental disorders.[1] Chronic diseases are known to be on the rise both globally and in Turkey. In 2021, 7 of the ten leading causes of death worldwide were from the chronic diseases group, and these 7 diseases were responsible for 68% of deaths in this group. The 3 leading chronic diseases in this group are ischemic heart disease, stroke, and COPD.[2] The leading cause of death in Turkey in 2023 was chronic diseases, with circulatory system diseases, cancers, and respiratory system diseases ranked as the most prevalent causes.[3] Chronic diseases have sociodemographic and genetic – non-modifiable – risk factors that include age, gender, race, ethnicity, education, income level, heredity and family history, and environmental, metabolic and behavioral – modifiable – risk factors. Modifiable risk factors are critical because they account for the preventable and protectable part of chronic diseases.[4] It is also known that behavioral risk factors are associated with an increased mortality risk.[5] In 2023, the age-standardized chronic disease mortality rate was 59.86 per 100,000 due to tobacco use, 11.72 due to alcohol use, 81.50 due to unhealthy diet, and 7.42 due to low physical activity levels, which are among the modifiable behavioral risk factors for chronic diseases in both sexes worldwide. In 2023 in Turkey, the age-standardized mortality rate due to chronic diseases was 350.11 per 100,000, and the rate due to tobacco use was 72.37 per 100,000, due to alcohol use 2.11, due to unhealthy diet 55.85 and due to low physical activity 11.78.[6] Understanding chronic diseases and their risk factors, identifying the current situation, and managing them appropriately remain key issues at local, national and global levels. Routine health checkups, one of the methods used for this purpose, have been found to increase the detection of chronic diseases.[7] The diagnostic benefit of cancer screenings and its impact on survival have also been shown.[8] In the current literature, evidence is limited regarding the extent to which individuals diagnosed with chronic diseases – who are in continuous contact with the healthcare system – adhere to lifestyle modifications and screening programs compared to their healthy peers. The use of Bayesian inference and effect size alongside traditional frequentist methods in this study provides a critical foundation for health policymakers. While frequentist methods test only for statistical significance (P-value), Bayesian methods frame uncertainty within a probabilistic context, and effect size quantifies the clinical and practical magnitude of the observed differences. Together, these approaches provide a more complete evidential basis for designing targeted behavioral interventions and prioritizing public health resources in chronic disease populations. This study aims to compare tobacco/alcohol use, fruit and vegetable consumption, physical activity, and participation in routine checkups and cancer screenings between individuals aged 40 and older with and without chronic diseases. The primary hypotheses of the study are: It is hypothesized that individuals with a chronic disease diagnosis will have higher participation rates in routine checkups and cancer screenings due to ongoing medical surveillance compared to their healthy peers. It is hypothesized that, despite this medical contact, individuals with chronic diseases will show similar or insufficient adherence to lifestyle modifications – including nutrition and physical activity – compared to healthy individuals. It is hypothesized that Bayesian analysis and effect size calculations will provide more in-depth evidence regarding behavioral risk patterns that are more suitable for policy decisions than frequentist analyses based solely on P-values. 2. Materials and methods This was a cross-sectional study. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were used in the reporting of this article.[9] The participant selection process and the transition from the initial database to the final analytic sample are illustrated in the flow diagram (Fig. 1). This study was conducted using secondary data obtained from the Turkish Statistical Institute (TurkStat; date: 08.09.2023). Since the study involved secondary analysis of existing data with no direct human contact and adhered to TurkStat’s institutional data use protocols, it was exempt from the requirement for local ethics committee approval. All data were provided in fully anonymized form, and no personally identifiable information was accessed or processed at any stage of the analysis, in accordance with applicable data protection principles. Flow diagram of participant selection and study sample based on STROBE guidelines. STROBE = Strengthening the Reporting of Observational Studies in Epidemiology, THS-2022 = Turkey Health Survey 2022. The data used in the study came from the Turkey Health Survey-2022 (THS-2022).[10] This study, the first of which was conducted in 2008, was intended to be representative of Turkey in general. The sample was selected using stratified 2-stage cluster sampling. The rural-urban distinction (settlements with a population of 20,000 and fewer are rural; 20,001 and more are urban) was used as a stratification criterion. The first-stage sampling unit consisted of clusters randomly selected in proportion to their size from clusters containing an average of 100 household addresses, and the second-stage sampling unit consisted of household addresses systematically selected at random from each selected cluster. Data for the THS-2022 were collected from 11,170 households through face-to-face interviews and 3 data collection forms (baseline characteristics of household members, 0 to 14 years questionnaire, and 15 years and older questionnaire). From the initial database of 29,761 individuals, 12,613 met the age inclusion criteria (40 years and older). After excluding 58 observations (0.46%) with missing values in key variables using complete case analysis (overall missingness rate < 5%), a final sample of 12,555 individuals was reached. Missing data were distributed across multiple variables, with no single variable exceeding 1% missingness; the pattern was considered to be missing completely at random. Included participants were citizens of the Republic of Turkey and diagnosed with at least one of the following conditions: myocardial infarction, coronary heart disease, hypertension, diabetes, stroke, and liver failure, which were considered relevant to the aim of the study and were among the chronic disease items included in the THS-2022 database. Data regarding chronic disease diagnosis were based on participants’ self-reports, in accordance with the TurkStat Health Survey protocol. For each chronic condition, participants were specifically asked whether they had experienced the relevant health problem within the last 12 months. A total of 29 variables were used in the study and the presence of chronic diseases was taken as a grouping variable. Analyses were performed using JASP (version 0.18.3.0) and IBM SPSS Statistics for Windows, version 25.0 (IBM Corp., Armonk). Numerical data were summarized using mean ± standard deviation and min-max values, and categorical data were summarized using numbers and percentages. Frequentist chi-square test, Phi, Cramer’s V, and Somers’ D (directional) effect sizes, and Bayesian chi-square test were used for bivariate analyses. Phi, which is calculated for nominal categorical variables and used in 2 × 2 contingency tables, can take values between −1 and +1, with 0 indicating no association. Similarly, Cramer’s V is calculated for nominal categorical variables in contingency tables larger than 2 × 2, and takes values between 0 and +1, with 0 indicating no association. Somers’ D, which is calculated for ordinal categorical variables, can take values between −1 and +1, with 0 indicating no association. These 3 coefficients were selected according to the measurement level and structure of the variables involved: Phi for nominal variables in 2 × 2 tables, Cramer’s V for nominal variables in larger contingency tables, and Somers’ D applied asymmetrically for ordinal variables, with the dependent variable specified according to the study’s aim. For the effect sizes, Cohen 0.10 small, 0.30 medium, and 0.50 large effect scales were used for chi-square test and correlation coefficients.[11] For the Bayesian chi-square test, the independent multinomial distribution was selected as the prior distribution, and seed and prior concentration were set to 1. Additionally, to control for the effects of demographic differences between groups, multivariable logistic regression models were employed. This allowed for the investigation of the association between chronic disease diagnosis and health behaviors while adjusting for key covariates (age, body mass index [BMI], sex, education level, marital status and employment status). Both frequentist and Bayesian logistic regression models were utilized to ensure the robustness of the results. Prior to the regression analyses, multicollinearity assumptions were checked to ensure the validity of the models. A P-value <.05 was considered to be statistically significant for the frequentist statistics. The predictive performance of the frequentist models was evaluated using Tjur’s coefficient of discrimination (Tjur R2) and classification accuracy. For the Bayesian logistic regression models, the relative evidence for the inclusion of each specific predictor was assessed using the inclusion Bayes factor (BFincl). Overall model fit comparisons against the base models were reported using the Bayes factor (BF10) and Bayesian R2. The interpretation of the Bayes factors was based on standard classification guidelines, where a BF between 1 and 3 indicates anecdotal evidence, 3 to 10 indicates moderate evidence, and >10 indicates strong evidence for the alternative hypothesis.[12,13] 3. Results 3.1. Sociodemographic characteristics of the study group The sociodemographic characteristics of the study group are shown in Table 1. The mean age was 56.26 ± 11.75 (40–102) for the whole group, 52.41 ± 10.25 (40–102) for those without chronic diseases, and 61.66 ± 11.59 (40–101) for those with chronic diseases. In the study group, 52% were female, 45.5% had completed primary school, 80.9% were married, and 66.6% were not working/working for free. The mean BMI of the study group was 27.92 ± 4.77 (13.84–66.33). The BMI of 42.8% of the group was overweight, 46.8% had good general health status, and 58.1% had no limitations due to health problems. Table 1Sociodemographic characteristics of the study group.Whole groupChronic disease statusFrequentist chi-square testEffect sizeBayesian chi-square testAbsentExistn%n%n%X2 P BF10 SexMale602348.03907a53.32116b40.5201.816<.0010.127(Phi)2.209 × 1042Female653252.03420a46.73112b59.5Education levelDid not finish school196215.6692a9.41270b24.3735.204<.001−0.175 (Somers’ D)1.750 × 10154Primary School571045.53223a44.02487b47.6Secondary School11949.5772a10.5422b8.1High School196915.71359a18.5610b11.7University150112.01112a15.2389b7.4Postgraduate2191.7169a2.350b1.0Marital statusSingle4563.6349a4.8107b2.0588.494<.0010.217 (Cramer’s V)1.483 × 10124Married10,15880.96177a84.33981b76.1Divorced6295.0426a5.8203b3.9Widowed131210.5375a5.1937b17.9Employment statusPaid employment419133.43273a44.7918b17.61008.419<.0010.283(Phi)1.703 × 10228Not working/working for free836466.64054a55.34310b82.4Work continuityPermanent work396291.0307291.289090.30.886.3460.014(Phi)0.088Temporary work3919.02958.8969.7Working modeFull time410194.23196a94.9905b91.813.755<.0010.056(Phi)43.47Part-time2525.8171a5.181b8.2BMIUnderweight (<18.5)1090.979a1.130b0.6601.395<.0010.166 (Somers’ D)9.776 × 10123Healthy weight (18.5–24.9)342027.22428a33.1992b19.0Overweight (25.0–29.9)537142.83213a43.92158b41.3Obese class 1 (30.0–34.9)269321.41270a17.31423b27.2Obese class 2 (35.0–39.9)7195.7265a3.6454b8.7Obese class 3 (40.0 or more)2431.972a1.0171b3.3General health statusPoor160712.8344a4.71263b24.22737.503<.0010.402 (Somers’ D)∞Average507240.42174a29.72898b55.4Good587646.84809a65.61067b20.4Restriction due to health problemNo restriction729958.15577a76.11722b32.92418.889<.001−0.398 (Somers’ D)∞Mild/moderate400831.91472a20.12536b48.5Severe12489.9278a3.8970b18.6a and b indicate the categories between which there is a difference. The results were analyzed using the frequentist chi-square test.BF = Bayes factor. Sociodemographic characteristics of the study group. a and b indicate the categories between which there is a difference. The results were analyzed using the frequentist chi-square test. BF = Bayes factor. 3.2. Tobacco and alcohol use, fruit and vegetable consumption, and physical activity habits of the study group Analyses of the group’s tobacco and alcohol use, fruit and vegetable consumption, and physical activity habits according to chronic disease status are presented in Table 2. There was a statistically significant difference between people without chronic diseases and those with chronic diseases in terms of tobacco use status, exposure to tobacco smoke, alcohol use status, frequency of alcohol consumption, daily activity status, walking at least 10 minutes a day, cycling at least 10 minutes a week, exercise at least 10 minutes a week, and time spent sitting/resting a day by both frequentist and Bayesian chi-square tests. The frequentist approach showed a difference in fruit consumption frequency (P = .048), but the Bayesian approach supported the null hypothesis of no difference (BF10 = 3.530 × 10-4). No difference was found between the frequentist and Bayesian approaches for vegetable/salad consumption frequency. The smallest and largest effect sizes were 0.007 and −0.180, respectively, and all were interpreted as small effect sizes. Table 2Tobacco and alcohol use, fruit and vegetable consumption, and physical activity habits of the study group.Chronic disease statusFrequentist chi-square testEffect sizeBayesian chi-square testAbsentExistn%n%X2 P BF10Tobacco use statusNo, never3783a51.63136b60.0281.146<.001−0.097 (Somers’ D)5.165 × 1056No, quit892a12.2896b17.1Yes, some days2743.71623.1Yes, every day2378a32.51034b19.8Exposure to tobacco smokeNever ever5049a68.93952b75.674.874<.001−0.073 (Somers’ D)6.843 × 1010At least once a week2843.91813.5Less than once a week4486.12815.4Every day, less than 1 h577a7.9287b5.5Every day, 1 h and more a day969a13.2527b10.1Alcohol use statusNo5887a80.34492b85.965.803<.0010.073(Phi)1.218 × 1013Yes1440a19.7736b14.1Alcohol consumption frequency*Do not use anymore455a31.6333b45.240.470<.001−0.094 (Somers’ D)3,04,550.794A few days a month694a48.2285b38.7A few days a week234a16.288b12.0Every day/almost every day574.0304.1Fruit consumption frequencyNever1111.5811.59.605.0480.012 (Somers’ D)3.530 × 10-4Less than once a week4516.23496.71–3 times a week1998a27.31304b24.94–6 times a week178924.4128424.6Every day297840.6221042.3Vegetable/salad consumption frequencyNever480.7430.85.519.2380.007 (Somers’ D)9.274 × 10-5Less than once a week2102.91683.21–3 times a week172323.5115722.14–6 times a week211428.9150628.8Every day323244.1235445.0Daily activity statusMostly sitting3332a45.53353b64.1486.302<.001−0.180 (Somers’ D)3.694 × 10102Mostly walking or work that requires moderate physical exertion3542a48.31772b33.9Mostly heavy work that requires physical strength453a6.2103b2.0Walking at least 10 min/dNo1583a21.61636b31.3149.691<.001−0.109(Phi)5.273 × 1030Yes5744a78.43592b68.7Cycling at least 10 min/wkNo7081a96.65130b98.124.624<.001−0.045(Phi)31,552.611Yes246a3.498b1.9Exercise at least 10 min/wkNo6948a94.85120b97.978.156<.001−0.079(Phi)1.483 × 1017Yes379a5.2108b2.1Time spent sitting/resting a dayLess than 4 h2578a35.21203b23.0396.549<.0010.130 (Somers’ D)2.764 × 1081More than 4 h to less than 6 h2370a32.31530b29.3More than 6 h to less than 8 h1320a18.01126b21.5More than 8 h1059a14.51369b26.2a and b indicate the categories between which there is a difference. The results were analyzed using the frequentist chi-square test.BF = Bayes factor.*Analyzed only for people who consume alcohol. Tobacco and alcohol use, fruit and vegetable consumption, and physical activity habits of the study group. a and b indicate the categories between which there is a difference. The results were analyzed using the frequentist chi-square test. BF = Bayes factor. Analyzed only for people who consume alcohol. People with chronic diseases had less tobacco use, less exposure to tobacco smoke, less alcohol use, and less physical activity. The fruit and vegetable/salad consumption frequencies were high and mostly similar in people with and without chronic diseases. It is important to note that due to the large sample size, both frequentist P-values and Bayesian Bayes factors yielded highly significant results. However, the calculated effect sizes ranged from 0.007 to −0.180, strictly indicating small effect sizes for all observed differences. 3.3. Participation in routine checkups and cancer screenings of the study group Analyses of the study group’s participation in routine checkups and cancer screenings are presented in Table 3. There was a statistically significant difference between people with and without chronic diseases for the variables last blood pressure measurement, last cholesterol measurement, last blood glucose measurement, last fecal occult blood test, last colonoscopy, last mammogram/breast X-ray, frequency of breast self-examination, and last cervical smear/pap test by both frequentist and Bayesian chi-square tests. The smallest and largest effect sizes were 0.027 and 0.262, respectively, and were interpreted as small and small-medium effects. Similar to lifestyle variables, despite the highly significant P-values and decisive Bayesian evidence driven by the large sample size, the objective effect sizes demonstrate that the magnitude of differences in routine checkup and screening participation between the groups remains modest. Table 3Participation in routine checkups and cancer screenings of the study group.Chronic disease statusFrequentist chi-square testEffect sizeBayesian chi-square testAbsentExistX2 P BF10n%n% Time of last blood pressure measurement by healthcare professional (HCP)Never ever815a11.1133b2.51318.656<.0010.262 (Somers’ D)2.381 × 10280More than 5 yr ago536a7.3133b2.5More than 3 yr to less than 5 yr673a9.2197b3.8More than 1 yr to less than 3 yr2182a29.8931b17.8Less than 1 yr3121a42.63834b73.3Time of last cholesterol measurement by HCPNever ever1225a16.7296b5.71177.807<.0010.240 (Somers’ D)6.301 × 10249More than 5 yr ago509a6.9147b2.8More than 3 yr to less than 5 yr691a9.4215b4.1More than 1 yr to less than 3 yr2058a28.11009b19.3Less than 1 yr2844a38.83561b68.1Time of last blood glucose level measurement by HCPNever ever1146a15.6269b5.11289.075<.0010.252 (Somers’ D)8.746 × 10273More than 5 yr ago524a7.2126b2.4More than 3 yr to less than 5 yr689a9.4204b3.9More than 1 yr to less than 3 yr2060a28.1944b18.1Less than 1 yr2908a39.73685b70.5Time of last fecal occult blood testNever ever4719a64.42918b55.8172.248<.0010.091(Somers’ D)1.065 × 1033More than 5 yr ago575a7.8358b6.8More than 2 yr to less than 5 yr73310.05159.9Less than 2 yr1300a17.71437b27.5Time of last colonoscopy testNever ever6518a89.04263b81.5136.330<.0010.146 (Somers’ D)1.069 × 1027More than 10 yr ago117a1.6140b2.7Less than 10 yr692a9.4825b15.8Time of last mammogram/breast X-ray*Never ever1733a50.71410b45.331.912<.0010.027 (Somers’ D)642.200More than 5 yr ago422a12.3486b15.6More than 2 yr to less than 5 yr524a15.3571b18.3Less than 2 yr74121.764520.7Frequency of breast self-examination*Never ever1493a43.71512b48.621.695<.001−0.042 (Somers’ D)3.990More than 3 mo72821.366321.3Quarterly374a10.9275b8.8Once a month825a24.1662b21.3Time of last cervical smear/pap test*Never ever173750.8165053.020.116<.001−0.032 (Somers’ D)35.017More than 5 yr ago453a13.2492b15.8Less than 5 yr1230a36.0970b31.2a and b indicate the categories between which there is a difference. The results were analyzed using the frequentist chi-square test.BF = Bayes factor, HCP = healthcare professional.*Analyzed only for female gender. Participation in routine checkups and cancer screenings of the study group. a and b indicate the categories between which there is a difference. The results were analyzed using the frequentist chi-square test. BF = Bayes factor, HCP = healthcare professional. Analyzed only for female gender. People with chronic diseases were more likely to attend both routine checkups and cancer screenings. However, the difference in the frequency of attending routine checkups for those with and without chronic diseases was more pronounced. 3.4. Predictors of chronic disease status: frequentist and Bayesian logistic regression analyses The results of the frequentist and Bayesian logistic regression analyses are presented in the supplementary material (Tables S1–S3, Supplemental Digital Content 1). The base models used in the analyses were constructed from sociodemographic variables (consisting of 6 variables in the first 2 models and 5 variables in the third). In the model concerning lifestyle variables for the whole group (Table S1, Supplemental Digital Content 1), both methods consistently identified age, BMI, sex, education level, marital status (widowed), employment status, and daily activity status as significant. However, tobacco use status, exercise at least 10 minutes a week, and time spent sitting/resting a day, which were found significant in the frequentist analysis, were not supported by the Bayesian analysis due to very low model inclusion factors (BFincl < 0.10). In the model incorporating routine checkups and certain cancer screenings (Table S2, Supplemental Digital Content 2), both methods found age, BMI, sex, education level, employment status, time of last blood pressure measurement by healthcare professional (HCP), and time of last blood glucose level measurement by HCP to be significant; time of last colonoscopy test showed significance only in the frequentist analysis (P = .015), whereas Bayesian evidence did not support this finding (BFincl = 0.048). In the cancer screening variables model for the female subgroup (Table S3, Supplemental Digital Content 3), age, BMI, and education level were significant in both methods, while frequency of breast self-examination and time of last cervical smear/pap test did not exhibit a significant relationship in either method. One of the most notable discrepancies between methods occurred in this model; while employment status and time of last mammogram/breast X-ray were found to be significant in the frequentist analysis but were not confirmed by the Bayesian analysis, marital status (widowed) lost its significance in the frequentist analysis (P = .060) but emerged as a significant risk factor in the Bayesian analysis (odds ratio [OR]: 1.458, 95% confidence interval [CI]: 1.005–2.075). In terms of performance comparison, while frequentist models exhibited higher Tjur R2 values (0.222; 0.276; 0.236) and accuracy rates (71–73%), Bayesian models produced more conservative R2 estimates (0.172; 0.224; 0.186). Nevertheless, Bayesian model comparison factors (BF10) indicated that the final model, obtained by including additional variables into the base model, provided a better fit to the data in all 3 analyses compared to the base model (2.50 × 106; 8.53 × 10183; 165.6). All 3 values substantially exceed the threshold of 10, which is conventionally considered strong evidence in favor of the alternative hypothesis (Table 4). Table 4Comparative summary of significant predictors of chronic disease status across frequentist and Bayesian logistic regression models (odds ratios).Variables (reference category)Table S1, Supplemental Digital Content 1: LifestyleTable S2, Supplemental Digital Content 2: Routine checkups and cancer screeningsTable S3, Supplemental Digital Content 3: Female cancer screeningsSociodemographic variables Age (continuous)1.063/1.0641.060/1.0611.073/1.073 BMI (continuous)1.104/1.1051.099/1.0991.095/1.095 Sex (female)1.309/1.220Non-sig./non-sig.– Education level (university, postgraduate for last model)*0.607/0.5690.460/0.4650.362/0.380 Marital status (widowed)1.554/1.537Non-sig./non-sig.Non-sig./1.458 Employment status (not working/working for free)1.332/1.3801.448/1.4611.200/non-sig.Lifestyle variables Tobacco use status (no, quit)1.391/non-sig.–– Exposure to tobacco smoke, alcohol, fruit, and vegetable consumption frequenciesNon-sig./non-sig.–– Daily activity status (mostly heavy work that requires physical strength)*0.557/0.546–– Exercise at least 10 min/wk (yes)0.718/non-sig.–– Time spent sitting/resting a day (>8 h)1.297/non-sig.––Routine checkups and some cancer screenings Time of last blood pressure measurement by HCP (<1 yr)*–3.309/3.343– Time of last blood glucose level measurement by HCP (<1 yr)–2.126/2.226– Time of last fecal occult blood test–Non-sig./non-sig.– Time of last colonoscopy test (<10 yr)–1.174/non-sig.–Female cancer screenings Time of last mammogram/breast X-ray (<2 yr)*––1.381/non-sig. Frequency of breast self-examination––Non-sig./non-sig. Time of last cervical smear/pap test––Non-sig./non-sig.Performance metrics Frequentist (Tjur R2)0.2220.2760.236 Bayesian (R2)0.1720.2240.186 Accuracy71.5%73.4%71.1% Bayes factor (BF10)2.50 × 1068.53 × 10183165.6BF = Bayes factor, BMI = body mass index, HCP = healthcare professional, OR = odds ratio.*Other categories also have a significant effect; however, for representation purposes, the OR values for the category with the strongest effect are presented in the table. Comparative summary of significant predictors of chronic disease status across frequentist and Bayesian logistic regression models (odds ratios). BF = Bayes factor, BMI = body mass index, HCP = healthcare professional, OR = odds ratio. Other categories also have a significant effect; however, for representation purposes, the OR values for the category with the strongest effect are presented in the table. The predictive performance and visual summary of the multivariable models (Tables S1–S3, Supplemental Digital Content 1) are presented in Figures 2–4. These forest plots display frequentist odds ratios and 95% confidence intervals. An asterisk (*) denotes variables supported by both frequentist and Bayesian evidence, while a dagger (†) indicates predictors significant only within the Bayesian framework. Unmarked variables represent significant frequentist results not confirmed by the Bayesian filter. Despite high statistical significance from the large sample size, the plots illustrate that objective effect sizes remain small to moderate. Forest plot of sociodemographic and lifestyle predictors for chronic disease status. BMI = body mass index. Forest plot of routine health checkups and general cancer screenings as predictors of chronic disease status. BMI = body mass index, HCP = healthcare professional. Forest plot of female-specific cancer screenings as predictors of chronic disease status. BMI = body mass index. 4. Discussion This study found that people with chronic diseases used tobacco and alcohol less than those without chronic diseases. Fruit and vegetable consumption was common and similar in people with and without chronic diseases. People with chronic diseases had lower levels of physical activity. The 2011 Chronic Diseases and Risk Factors Study in Turkey, which included 18,477 people, found that 30% of the population smoked tobacco, 13% consumed alcohol, 4% did not consume fruits and vegetables, and 13.1% consumed 5 or more portions of fruits and vegetables daily. The results of the same study showed that 55% of men and 69% of women had inadequate physical activity levels, and the proportion of people doing adequate and moderate physical activity decreased with increasing age.[14] There are studies in the literature showing that tobacco use was common, alcohol use was limited in amount and common, fruit and vegetable consumption and physical activity were low in people with chronic diseases.[15-19] This study differs from others in terms of tobacco and alcohol use and fruit and vegetable consumption. The results of this study may indicate that people with chronic diseases show a healthier tendency to be aware of their disease and to limit their tobacco and alcohol use, increasing their fruit and vegetable consumption or maintaining their pre-disease levels. This behavioral shift may be driven by post-diagnosis medical counseling, disease-specific health literacy, and the immediate fear of disease exacerbation, prompting patients to consciously restrict acute risk factors like smoking and drinking. In line with the literature, the lower levels of physical activity found in this study may indicate that people with chronic diseases lead more sedentary lives, away from activity. Rather than a mere lack of awareness, this physical inactivity may often be directly secondary to the disease process itself – such as chronic pain, disease-induced fatigue, reduced mobility, or fear of injury – compounded by the fact that the mean age of people with chronic diseases was higher than that of those without chronic diseases. Various results from studies in the literature suggest that people maintain behavioral risk factors associated with chronic diseases. This may be related to the characteristics of the study setting and population, as well as cultural and social norms, and the lack of awareness among people with chronic diseases. This study found that people with chronic diseases were more likely to participate in routine checkups and cancer screenings, with a larger effect size for their participation in routine checkups. However, participation in cancer screenings was similar in both groups and had a small effect size. The measurement most frequently missed by the group was cholesterol, followed by blood glucose. The most commonly missed cancer screening tests were colonoscopy, fecal occult blood test, and cervical smear/pap test, respectively. In this study, the observation that people with chronic diseases were more likely to be included in routine checkups may be related to the fact that these people visit health facilities more often due to the treatment and control of the diseases. It is noteworthy, however, that although people with chronic diseases interact more often with health professionals during the measurements, this interaction does not lead to an increase in cancer screenings. This discrepancy may point to a phenomenon of competing clinical priorities during medical visits; physicians managing chronic conditions often focus heavily on immediate disease control within limited appointment times, which may inadvertently overshadow preventive cancer screenings. A study of a group of middle-aged people with diabetes found high rates of participation in health checks.[20] In a study assessing participation in cancer screenings, low rates were found.[21] Routine checkups and cancer screenings are essential for early diagnosis and disease management. Each country may offer different programs for routine checkups and cancer screenings as part of primary health care. Studies show that attending routine checkups is associated with the perceived risk to one’s health and that there may be barriers to this, such as lack of time, procrastination/avoidance, and poor socioeconomic status.[22-24] Lack of knowledge and awareness, lack of social support, distance, fear, and financial reasons underlie delays in cancer screenings.[25,26] It seems important to address the barriers to participation in routine checkups and cancer screenings and to increase their coverage. When extending our analyses to multivariable logistic regression models, the robustness of core sociodemographic variables and critical clinical measurements was consistently confirmed by both frequentist and Bayesian approaches. However, the logistic regression models also highlighted critical methodological divergences. In our analyses, while the frequentist and Bayesian chi-square tests were mostly parallel for bivariate associations, the Bayesian approach acted as a more conservative filter in multivariable settings. For instance, lifestyle variables such as tobacco use status, exercise, and daily sitting time, as well as specific clinical procedures like colonoscopy history, yielded significant P-values in the frequentist models. Conversely, Bayesian evidence did not support their independent predictive value. This indicates that while these factors may show superficial bivariate associations, their independent contribution to predicting chronic disease status appears limited when adjusting for stronger covariates like age and BMI. A critical methodological consideration in interpreting all these findings – and a notable contribution of this study to the current literature – is the distinction between statistical significance and practical significance. Given the extensive sample size of our study, even minuscule deviations from the null hypothesis generated highly significant P-values in the frequentist approach and decisive Bayes factors in the Bayesian approach. This dual-method comparison highlights a potential caveat in large-scale epidemiological studies: while Bayesian methods are generally more conservative, both paradigms remain highly vulnerable to massive datasets, leading to highly significant statistical evidence for almost all variables. A key contribution of this study is the empirical demonstration of why effect sizes should be carefully considered alongside traditional significance testing in large cohorts. While P-values and Bayes factors are sensitive to sample size, effect sizes provide a crucial complementary perspective by directly quantifying the magnitude of an association. Although effect size estimates are not immune to limitations – such as sensitivity to measurement error and the requirement of adequate sample sizes to yield precise confidence intervals – they offer a stable and clinically informative metric for evaluating real-world relevance in large-scale cohorts. In our study, these objective effect sizes primarily ranged from very small to moderate. This indicates that although the differences in lifestyle habits between people with and without chronic diseases are statistically significant, the small effect sizes show that these differences are minimal in practice. By demonstrating that decisive statistical evidence often masks trivial real-world differences, this study highlights the value of a broader perspective in public health research: complementing dichotomous hypothesis testing with the routine reporting of effect sizes. There are studies in the literature that have emphasized similar methodological concerns regarding sample size and effect size reporting.[27,28] 4.1. Study highlights and limitations In this study, a large and representative sample was utilized, supporting the generalizability of the findings; furthermore, as THS-2022 was the most recent wave of the survey available at the time this study was conducted, the findings reflect up-to-date individual-level behavioral and screening patterns. Nevertheless, as the data are derived from Turkey, caution should be exercised when generalizing the findings to populations with different healthcare systems or cultural contexts. No ecological bias was present, as individual-level data were used. Effect size was calculated in the analysis phase of the study, in addition to using frequentist and Bayesian methods. This combined approach provides a more balanced evaluation of the associations, reducing reliance on sample-size-sensitive P-values. These characteristics were the strengths of this study. However, there were limitations in this study inherent to collecting data by questionnaire. Self-reporting can be prone to recall and social desirability biases, meaning participants might slightly overstate healthy behaviors or understate risks. Similarly, chronic disease status itself was based on self-report rather than clinical or laboratory confirmation, which may introduce misclassification bias. Such non-differential misclassification would generally be expected to attenuate the observed associations toward the null, suggesting that the true differences between groups could be somewhat larger than reported here. Additionally, due to the cross-sectional design, causal relationships cannot be definitively established; thus, the results represent associations rather than causations. 5. Conclusion The study evaluated data from 12,555 people with at least one of the following conditions: myocardial infarction, coronary heart disease, hypertension, diabetes, stroke, or liver failure. Overall, the findings indicate that people with chronic diseases tend to use less tobacco and alcohol, do less physical activity, and have more routine checkups and cancer screenings compared to those without chronic conditions, while fruit and vegetable consumption was similar between the 2 groups. Crucially, evaluating the absolute magnitude of these differences suggests that while these behavioral shifts are statistically detectable, the real-world changes following a chronic disease diagnosis are often modest. Based on these findings, the following recommendations can be considered for practice and future research: First, instead of generic advice, public health interventions could benefit from offering targeted, low-impact physical activity programs tailored to the limitations of people with chronic diseases. Second, to support participation in cancer screenings without being overshadowed by routine disease management, healthcare policies may consider integrating automated alerts into electronic records or utilizing dedicated nurses. Finally, future health research and policy-making may benefit from placing more emphasis on the practical magnitude of behavioral differences – by routinely reporting effect sizes – alongside statistical significance. Author contributions Conceptualization: Elif Nur Yildirim-Ozturk. Data curation: Elif Nur Yildirim-Ozturk. Formal analysis: Elif Nur Yildirim-Ozturk. Investigation: Elif Nur Yildirim-Ozturk. Methodology: Elif Nur Yildirim-Ozturk, Mehmet Uyar. Project administration: Elif Nur Yildirim-Ozturk. Resources: Elif Nur Yildirim-Ozturk. Software: Elif Nur Yildirim-Ozturk, Mehmet Uyar. Supervision: Elif Nur Yildirim-Ozturk, Mehmet Uyar. Validation: Mehmet Uyar. Visualization: Elif Nur Yildirim-Ozturk. Writing – original draft: Elif Nur Yildirim-Ozturk. Writing – review & editing: Elif Nur Yildirim-Ozturk, Mehmet Uyar.