EXPLORING VITAMIN D-RELATED KAP INFLUENCING FACTORS IN MIDDLE-AGED AND ELDERLY PATIENTS WITH ANKLE FRACTURES: A LOGISTIC REGRESSION AND DECISION TREE MODEL-BASED STUDY

1. Introduction

China rapidly approaches an aging society, with a significant increase in the elderly population.¹ Ankle fractures are common, accounting for about 3.92% of all fractures² and 7.6% of adult fractures.³ A corresponding increase in ankle fractures is observed among middle-aged and elderly people. Calcium and Vitamin D (VD) deficiencies are the key factors affecting post-traumatic fracture healing.^4,5 Moreover, the middle-aged and elderly populations often suffer from decreased bone mass,⁶ as well as insufficient or deficient VD levels.⁷ Additionally, VD-related nutritional knowledge, attitude, and practice (KAP) tend to be at a generally low level.⁸ VD, an essential physiological hormone in the human body, is of great significance for bone regulation. In middle-aged and elderly populations, the synthesis of VD by the skin decreases, along with a gradual decline in the ability of the kidney to further convert it. In addition, limited outdoor activities, poor eating habits, and other problems contribute to the worsening of VD insufficiency or deficiency. At the same time, VD deficiency is a multifactorial phenomenon affected by a series of environmental and social factors. The status of VD-related KAP of middle-aged and elderly populations also belongs to the category of influencing factors. Current research on KAP-influencing factors predominantly uses logistic regression models, whereas decision trees visually represent the importance of influencing factors. Combining these models has been found to enhance analytical efficiency more than using them individually.^9,10

Consequently, this study applied binary logistic regression and decision tree models to investigate the factors affecting VD-related KAP among patients. The objective was to offer key insights for developing specific intervention strategies. The study may provide a scientific foundation for improving VD levels in elderly patients with ankle fractures, thereby facilitating a reduction in their recovery time. This, in turn, may expedite the surgical rehabilitation process, reduce hospitalization expenses, and enhance patient satisfaction. The implications of this study may also contribute to refining healthcare approaches and outcomes for this specific patient group, particularly in geriatric orthopedics.

2. Subjects and Methods

2.1. Research subjects

This study conducted a questionnaire survey on middle-aged and elderly patients with ankle fractures who met the inclusion criteria in the orthopedic ward of a tertiary A-grade hospital in Bincheng District, Binzhou City, Shandong, from 1 January to 30 August 2023. The inclusion criteria for the patients were as follows: admitted for ankle fracture and underwent surgical treatment; aged $\ge 45$ years; normal daily living and activity abilities prior to the fracture; normal mental, psychological, and cognitive abilities, with normal expression and comprehension skills; able to understand Mandarin well and recognize Chinese characters either independently or with the aid of glasses. The exclusion criteria were pathological fractures, history of ipsilateral ankle fracture, recent history of other surgeries, multiple fractures combined with severe organ damage, noncooperation in the survey, history or current mental illness, consciousness disorders, and communication barriers. The study used a coin-toss method for random sampling of patients who met the criteria (heads indicated selection for the questionnaire survey).

Due to the minimal risk posed by our study procedures, only verbal consent was obtained from the patients and their families after explaining the objectives of and reasons for the study and assuring them of their privacy and a promise to keep the information confidential. This study was approved by the Ethical Committee of the Binzhou Medical University Hospital (Ethics ID [2023]KT-94 and [2023]LW-106).

2.2. Survey method

2.2.1. Survey tool

The study utilized the Chinese version of the Vitamin D-related KAP scale, originally developed by Amiri et al.¹¹ and translated into Chinese by Liang et al.¹² The translated scale has a total Cronbach’s alpha coefficient of 0.795 and a content validity S-CVI/UA of 0.947. The scale comprises 38 items across three dimensions: Nutrition Knowledge (16 items), Attitude (12 items), and Behavior (10 items). For the Knowledge dimension, the possible answers are “Yes, No, I don’t know.” Except for items 13, 14, 15, and 16, which are scored as 0, 2, and 1, respectively, other items are scored as 2, 0, and 1, with a score range of 0–32. The Attitude dimension is based on a Likert scale of 5, ranging from “Strongly Disagree” to “Strongly Agree,” with corresponding scores from 1 to 5, totaling 12–60 points. The Behavior dimension is also based on a Likert scale of 5, ranging from “Never” to “Always.” Items 29, 30, 31, 32, 34, and 36 are scored positively from 1 to 5, while others are scored inversely, with a total score range of 10–50. Scores in each dimension are proportionally converted into a 0–100 scale. In this study, the scale’s Cronbach’s alpha coefficient was 0.876.

2.2.2. Data collection

Data was collected from questionnaires filled out by the study subjects, which were available in both electronic and paper formats. The subjects could choose the format most convenient for them.

2.3. Quality control

All surveyors were systematically trained. With the cooperation of medical staff in relevant departments, the purpose, significance, and method of filling out the questionnaire were explained to the selected middle-aged and elderly patients with ankle fractures. The patients independently completed the questionnaire, which was collected on the spot. Immediate verification was conducted upon collection, with subjects asked to fill in or correct any missing or erroneous responses. After distributing the questionnaires each day, surveyors remained in the ward, making rounds intermittently to answer questions and collect completed questionnaires, leaving the ward only after collecting all the questionnaires distributed that day.

2.4. Model principles

Logistic Regression: Also known as logistic regression analysis, it is used to analyze the relationship between the probability of a certain type of event occurring and independent variables. This method is particularly suitable for data where the dependent variable is binary. The independent variables in the model can be qualitative discrete values or quantitative observed values.¹³ The regression model assumes that the probability distribution of sample observations on the dependent variable is S-shaped. Since it is a nonlinear model, maximum likelihood estimation is often used in parameter estimation to maximize the probability of the number of observations of the dependent variable, and then the best estimate of the parameters of the independent variable is obtained. The binary dependent variable logistic regression model was as follows :

There is a linear relationship between the dependent and independent variables $x_k$, The significance test of regression model included overall model test and individual parameter test, −2 log-likelihood test and Hosmer–Lemeshow test for overall test by likelihood value, etc. The significance test of individual parameters was mainly Wald test and Score test. It is relatively robust in modeling, providing a good overall fit and less prone to overfitting. It also allows for the exploration of the association between independent and dependent variables after correcting for confounding variables presenting results in terms of odds ratios and confidence intervals.¹⁴

Decision Tree CHAID Algorithm: A decision tree is a data mining method known for its intuitiveness and ease of understanding, used for exploring intrinsic data patterns and predicting new data object classifications. It encompasses various algorithms, including the Chi-squared Automatic Interaction Detector (CHAID). This algorithm can handle categorical and numerical input and output variables, build multi-branch trees, and determine the best grouping variables and split points from a statistical significance testing perspective.¹⁵ It is unaffected by the interrelationships between independent variables and can reveal interactions among them.¹⁶

2.5. Statistical analysis

All KAP dimension scores were calculated on a percentage scale: Standard score$=$Average score/Total $\mathrm{\text{score}}\times 100\%$.^17,18 Scores >85 are considered good, <60 are poor, and those between 60–85 are average.¹⁹

The scores of the Chinese version of the D-KAP-38 scale are quantitative data, presented as mean±standard deviation. T-tests and Chi-squared tests were used to analyze nutritional knowledge, attitudes, and behaviors of subjects with different ages and education levels. Logistic regression models and the decision tree CHAID algorithm were employed to examine whether subjects’ KAP scores were $\ge 60$ points as the dependent or target variable. Factors deemed statistically significant in univariate analysis were used as independent or input variables to explore the main factors and their effects on Vitamin D nutritional KAP in the subjects. All analyses were conducted in SPSS 25.0, with a test level of $\alpha =0.05$.

3. Results

3.1. Basic information

A total of 570 questionnaires were distributed, of which 5 were invalidated due to ink leakage or contamination by food spills, resulting in 565 valid questionnaires with an effective response rate of 99.12%. The study included 565 participants, with 227 males (40.18%) and 338 females (59.82%). The age distribution was as follows: 310 participants aged 45–59 years (54.87%), 250 participants aged 60–74 years (44.25%), 4 participants aged 75–89 years (0.71%), and 1 participant aged 90 years or above (0.18%).

3.2. Overall status of VD-related KAP in middle-aged and elderly patients with ankle fractures

Overall status of VD-related KAP in middle-aged and elderly patients with ankle fractures is provided in Table 1.

3.3. Analysis of factors influencing VD-related KAP in middle-aged and elderly patients with ankle fractures

Univariate analysis was conducted to determine the relationship between various factors and the VD-related KAP of patients. To further clarify the impact of these factors on patients’ VD-related KAP, analyses using the logistic regression model and the decision tree CHAID algorithm were performed. Detailed variable assignments are given in Table 2.

3.4. Logistic regression analysis of factors influencing VD-related KAP in middle-aged and elderly patients with ankle fractures

The analysis used whether the VD-related KAP scores were $\ge 60$ points and whether the scores in the three dimensions of knowledge, attitude, and behavior (Y, Y1, Y2, Y3) were $\ge 60$ as dependent variables. A value of 0 was assigned for ‘no’ and 1 for ‘yes,’ with other variables as independent variables. A binary logistic regression analysis was established, with results presented in Tables 3–6. The results indicated that the age of patients, the education level of the spouse, and the occupation of the spouse were the main factors affecting Vitamin D-related KAP, with the patient’s age group ($\mathrm{\text{OR}}=0.186$) and spouse’s education level ($\mathrm{\text{OR}}=0.336$) being protective factors. The patient’s age group and educational level were the main factors influencing Vitamin D-related knowledge, with the patient’s age group ($\mathrm{\text{OR}}=0.311$) as a protective factor. The patient’s age group ($\mathrm{\text{OR}}=0.020$) and spouse’s education level ($\mathrm{\text{OR}}=0.312$) were the main factors influencing Vitamin D-related attitudes, both being protective factors. The patient’s age group, spouse’s education level, and average monthly household dietary expenditure were the main factors influencing Vitamin D-related behavior, with the patient’s age group ($\mathrm{\text{OR}}=0.281$) and spouse’s education level ($\mathrm{\text{OR}}=0.420$) being protective factors.

3.5. CHAID algorithm analysis of decision tree for factors influencing VD-related KAP in middle-aged and elderly patients with ankle fractures

The analysis used whether the patients’ VD-related KAP scores were $\ge 60$ points and whether the scores in the three dimensions of knowledge, attitude, and behavior (Y, Y1, Y2, Y3) were $\ge 60$ as target variables, with other variables as input variables. The CHAID algorithm in the decision tree was used for analysis, setting the maximum depth of the tree to 3 layers, with a minimum of 100 cases for parent nodes and 50 for child nodes. The constructed decision tree model analysis is shown in Figs. 1–4. The results showed that the spouse’s education level, the patient’s age group, and the intensity of housing illumination were influencing factors for Vitamin D-related KAP. The top layer was the spouse’s education level, indicating its highest correlation with KAP. Patient’s education level, age group, and spouse’s occupation were influencing factors for Vitamin D-related knowledge, with the patient’s education level being the top layer, indicating its highest correlation with knowledge. Spouse’s education level, patient’s age group, and residential room lighting condition were influencing factors for Vitamin D-related attitude, with the spouse’s education level being the top layer, indicating its highest correlation with attitude. Spouse’s education level, patient’s age group, and housing illumination intensity were influencing factors for Vitamin D-related behavior, with the spouse’s education level being the top layer, indicating its highest correlation with behavior.

4. Discussion

4.1. Overall status of VD-related KAP in middle-aged and elderly patients with ankle fractures

This study, spanning 8 months, covered all four seasons in the local area. The average score of VD-related KAP among middle-aged and elderly patients with ankle fractures was ($67.08\pm 14.38$), which was considered moderate. This finding aligned with the results of a nutritional KAP survey among mothers of infants and young children in Shihezi.¹⁹ However, the knowledge dimension scored the lowest on average, inconsistent with previous studies^12,20 that reported higher knowledge and attitude scores compared with behavior scores. This discrepancy could be attributed to the different scales used in each study. Additionally, previous research mostly involved younger and middle-aged adults, while this study involved middle-aged and elderly individuals, who might have lower levels of education and knowledge.

4.2. Analysis of influencing factors for VD-related KAP in middle-aged and elderly patients with ankle fractures

The logistic regression analysis showed that the patient’s age group [odds ratio $\mathrm{\text{(OR)}}=0.186$, $P<0.001$] and spouse’s education level ($\mathrm{\text{OR}}=0.336$, P< 0.001) were protective factors for VD-related KAP. The older the patient and the higher the education level of the spouse, the better the overall status of VD-related KAP. The decision tree model analysis revealed that the spouse’s education level, patient’s age group, and housing illumination intensity were significant factors affecting VD-related KAP. The spouse’s education level, positioned at the top of the decision tree, indicated its highest correlation with KAP. The logistic regression model identified different meaningful variables compared with those entering the decision tree model nodes. It reflected the dependence among the patient’s age group, spouse’s education level, and spouse’s occupation. However, it did not emphasize the role of housing illumination intensity. The possible interaction effects between housing illumination intensity and spouse’s education level or patient’s age group could explain this. Although the logistic regression model demonstrated the interrelationships between VD-related KAP and various variables, it did not visually represent the importance of each influencing factor. Guo et al.²¹ found that the decision tree model analysis of factors influencing medication adherence in patients with severe mental disorders had a higher accuracy rate (80.9%) compared with the logistic regression model (71.4%). This suggested that the decision tree model analysis could, to some extent, circumvent the limitations of logistic regression models. It helped clarify the importance of input variables in relation to the target variable and visually presented it in a tree diagram. Applying the decision tree model could effectively complement the logistic regression model by explaining the importance of various factors to the outcome variable and presenting the analysis process more clearly.⁹ The decision tree analysis revealed that patients aged 60–74 years, whose spouses had a junior high school or high school/junior college education level, had a higher likelihood of scoring below 60 in VD-related KAP, indicating a generally poor level. Patients with an elementary school education or lower, combined with nonaverage housing illumination intensity, had a high proportion (97.7% or 216/221) of scores $\ge 60$, indicating a generally good level. Furthermore, neither analysis included the patient’s education level, suggesting that it might not have influenced the outcome in the overall situation of VD-related KAP. Instead, a lower education level in spouses seemed beneficial for patients’ VD-related KAP status. The findings indicated that relying solely on educational attainment as an indicator to evaluate the true cultural literacy of patient populations might be overly simplistic. Future studies should aim to expand the assessment criteria to include factors such as social practice and theoretical application for a more holistic understanding.

4.2.1. Analysis of factors influencing VD-related knowledge in middle-aged and elderly patients with ankle fractures

The logistic regression analysis revealed that the patient’s age group ($\mathrm{\text{OR}}=0.311$, $P<0.001$) was a protective factor for VD-related knowledge. The older the patient, the better the VD-related knowledge. The decision tree model analysis identified the patient’s education level, age group, and spouse’s occupation as significant factors influencing VD-related knowledge. The patient’s education level was positioned at the top of the decision tree, with the tree’s growth structured around stratification by the spouse’s education level. Additionally, the patient’s age group and spouse’s occupation contributed further explanatory power. Logistic regression is a classic multifactorial analysis model that quantifies the relationship between influencing factors and the dependent variable after controlling for other factors. In contrast, the decision tree model clearly and intuitively displays the analysis process and outcomes, emphasizing the significance of influencing factors, dividing subgroups with different characteristics, and facilitating the rapid identification of people with different characteristics to adopt appropriate measures. This approach provides more practical guidance than logistic regression, which only identifies the influencing factors.^22,23,24

This study showed that among patients with an education level of bachelor’s/associate degree or master’s degree or higher, 62.1% (36/58) scored $\ge 60$ points in VD-related knowledge. Patients with an education level of elementary school or below and whose spouse’s occupation was corporate staff had the lowest scores, with only 4.5% (3/67) scoring $\ge 60$ points. The higher the patient’s education level, the better the knowledge score, which was consistent with the findings of Liu et al.²⁵

4.2.2. Analysis of factors influencing attitudes toward VD in middle-aged and elderly patients with ankle fractures

The logistic regression analysis showed that the patient’s age group ($\mathrm{\text{OR}}=0.020$, $P<0.001$) and spouse’s education level ($\mathrm{\text{OR}}=0.312$, $P=0.007$) were protective factors for attitudes related to VD. The older the patients and the higher the education level of their spouses, the better the attitudes toward VD. The decision tree model analysis revealed that the spouse’s education level, patient’s age group, and housing illumination intensity significantly influenced patients’ attitudes toward VD. The spouse’s education level was positioned at the top of the decision tree, indicating that the growth of the tree primarily depended on the division by the spouse’s education level, with the patient’s age group and housing illumination intensity providing additional explanatory power. Logistic regression is used to explore influencing factors, revealing dependencies between independent and dependent variables. However, the interaction effects in the regression model may not be readily apparent when many independent variables exist.^26,27 The decision tree model can compensate for these shortcomings in logistic regression and has been widely applied in medical and health-related research.^28,29,30

The results indicated that patients whose spouses had an education level of elementary school/below or master’s degree and above, combined with nonaverage housing illumination intensity, scored the best in attitudes related to VD, with 99.1% (220/222) scoring $\ge 60$. Research suggests that patients’ attitudes toward VD are not influenced by their education level but by the education level of their spouses. It is possible that a higher education level of the spouse leads to increased positive family support for the patient. Subsequently, social support factors can be refined and included in the investigation, leading to more interesting and meaningful results. Moreover, the average score for attitudes is high ($72.99\pm 18.77$), indicating a generally positive attitude toward VD among patients, similar to the findings of Wu and others.³¹

4.2.3. Analysis of factors influencing behavior related to VD in middle-aged and elderly patients with ankle fractures

The logistic regression analysis demonstrated that the patient’s age group ($\mathrm{\text{OR}}=0.281$, $P<0.001$) and spouse’s education level ($\mathrm{\text{OR}}=0.420$, $P<0.001$) were protective factors for behavior related to VD. The older the patient and the higher the education level of the spouse, the better the behavior related to VD. The decision tree model analysis identified spouse’s education level, patient’s age group, and residential room lighting condition as significant factors influencing VD-related behavior. In the decision tree model, the spouse’s education level was at the top, indicating that the tree’s growth was primarily based on the spouse’s education level. The patient’s age group and residential room lighting condition offered supplementary explanations. Both logistic regression and decision tree models highlighted the importance of the patient’s age group and spouse’s education level as key influencers of VD-related behavior. The average monthly household dietary expenditure, a significant variable in logistic regression analysis, did not enter the decision tree model. This discrepancy might be related to the sample size of this study, affected by the settings of the sample size at each node and the depth limitations of the decision tree. This might have prevented the impact of household dietary expenditure from being adequately represented in the decision tree analysis. Alternatively, in the decision tree analysis, even if average monthly household dietary expenditure had some impact on VD-related behavior, its influence might have been weaker compared with other variables, leading to its identification as a confounding factor and exclusion from the final analysis.^32,33

4.3. Limitations

As a single-center study, this study had certain limitations, particularly regarding the representativeness of the sample. Future research should involve multi-city collaborations, incorporating numerous medical institutions and community health service organizations across various levels, focusing on middle-aged and elderly patients with ankle fractures. This approach will expand the sample size and potentially include a broader range of influencing factors. Also, relevant clinical objective indicators, such as blood calcium concentration, serum 25-hydroxyvitamin D levels, and so forth, can be incorporated to further explore the factors influencing VD-related KAP in middle-aged and elderly patients with ankle fractures. Such comprehensive research can provide more accurate results and offer a reference for developing a nutrition model for patients with fractures, which is suitable for the national conditions of China.

4.4. Conclusion

VD-related KAP in middle-aged and elderly patients with ankle fractures is influenced by various factors. Enhancing family support and self-cultivation to prevent VD deficiency is crucial for promoting patient recovery and healing. Both logistic regression and decision tree models can identify the factors influencing VD-related KAP. Logistic regression quantifies the relationship between KAP and influencing factors, whereas the decision tree model intuitively reflects the impact and importance of various factors on KAP. It also divides subgroups and illustrates complex interactions between independent variables. These two methods can be used in conjunction, complementing each other, to elucidate the correlations between variables further.

Acknowledgments

We would like to thank all participating patients and their families for their enthusiastic cooperation. We also acknowledge the assistance of the nurses and doctors of the hospital in Shandong, China.

Declaration of Conflicting Interests

The authors declare that there is no conflict of interest.

Funding

This study was funded by the 2022 Binzhou Medical University Hospital Nursing “New Seed Project” (grant number BYFYHL-XM202207) and the 2022 Binzhou Medical College campus-level research project (grant number BY2022SK01).

Supplementary Data

$〈〈$D-KAP-38 questionnaire$〉〉$

Nutrition knowledge

1.	People, who work indoors, are at high risk of vitamin D deficiency.
2.	Vitamin D intake more than dietary recommendations could be harmful.
3.	Elderly people are at high risk of vitamin D deficiency.
4.	Inappropriate dietary intakes are related to vitamin D deficiency.
5.	Vitamin D supplement intake requirements, differ for different age groups.
6.	Pregnant and lactating women are at high risk of vitamin D deficiency.
7.	Most of the vitamin D required is produced when the skin is directly exposed to the sun.
8.	Currently, vitamin D deficiency is one of the most important health issues in our country.
9.	Bone pain and fatigue are among the vitamin D deficiency symptoms.
10.	Vitamin D supplement intake requirements, differ in various seasons of the year.
11.	Both men and women are at risk of vitamin D deficiency.
12.	Fatty fishes are one of the main dietary sources of vitamin D.
13.	Dairy products are one of the main dietary sources of vitamin D.
14.	Eggs are one of the main dietary sources of vitamin D.
15.	Meat and poultry are the main dietary sources of vitamin D.
16.	Fruits are one of the main dietary sources of vitamin D.

Attitude

17.	Urbanization prevents sun exposure and production of required vitamin D.
18.	A shortage of public places for outdoor activities prevents the sun exposure required for production of vitamin D.
19.	Full time indoor occupation prevents the sun exposure required for production of vitamin D.
20.	Inefficient education regarding benefits of sun exposure prevents production of required vitamin D through sun exposure.
21.	The undesirable taste of sea foods for Iranians is one of the barriers to their consumption of dietary sources of vitamin D.
22.	In vitamin D deficiency, supplement intake is more effective compared to dietary intake and sun exposure.
23.	Taking vitamin D supplement, unless recommended by physicians is wrong.
24.	Unwillingness of individuals to take vitamin D supplement is one of the barriers of providing this nutrient.
25.	Taking supplements is necessary for treatment of vitamin D deficiency but not for its prevention.
26.	Permanent using of sunscreens on face, neck, and hands prevents the sun exposure required for production of vitamin D.
27.	Taking supplement is only necessary in case of lack of exposure to sunlight.
28.	A high expense of dietary sources of vitamin D is one of the barriers of providing this nutrient.

Practice

29.	For sufficient exposure to sunlight I regularly engage in outdoor physical activities.
30.	To be vitamin D sufficient, I consume fortified milk.
31.	In order to be vitamin D sufficient, I consume fish at least twice a week.
32.	For sufficient exposure to sunlight I walk outdoors daily.
33.	I use caps/hats to avoid severe sun exposure.
34.	To be vitamin D sufficient, I take vitamin D supplements.
35.	I use sunscreen on my hands.
36.	During the day I am directly exposed to sunlight (outdoors).
37.	During the day I am indirectly exposed to sunlight (through glass).
38.	I use sunscreen on my face.

ORCID

Yu Liang  https://orcid.org/0000-0003-3373-111X

Guanglei Zhang  https://orcid.org/0000-0002-3792-0500