Background

Chronic diseases are a group of diseases with insidious onset, long duration, and persistent symptoms [3, 4], ultimately resulting in an increased incidence of adverse drug reactions (ADRs) [5].

Our previous study has discovered a negative correlation between polypharmacy behaviors and shared decision making (SDM) among chronic disease patients in community [6]. This study was conducted to further explore the factors influencing chronic disease patients’ participation in SDM on medication based on a similar cohort. SDM refers to a collaborative process in which patients and their doctors discuss the pros and cons of various medical regimens, consider patients’ values and preferences, and finally make medical decision together [7]. Furthermore, participants in SDM may also involve multiple medical staff and patients’ social networks [8]. However, some evidences have showed that most patients reported lower levels of SDM and retained doctor-led views on decision-making, especially engrained in the elderly [9, 10]. Although low levels of SDM among elderly patients with chronic diseases have garnered attention from researchers, several studies focused on implementing programs to enhance patients’ experience in the SDM process, rather than figuring out what factors stimulate patients to participate in this process [11,12,13]. Considering that patients' participation is the foundation of SDM, we believe that the first step is to identify the factors that promote patients to participate in SDM, and then take corresponding measures to improve the participation in SDM among elderly patients with chronic diseases.

China has the largest elderly population in the world, facing the critical challenge of chronic diseases [15]. Current studies indicated that the chronic disease population in China showed improvements in health knowledge, medication compliances, and health habits [3, 6]. Firstly, we categorized Hubei Province into urban and rural areas, and then randomly selected 2 cities in each of category. Wuhan and Yichang were selected as sample urban areas, while Zhijiang and Qianjiang were selected as sample rural areas. Secondly, we followed the same simple randomization process to choose 3 districts in each of the four selected cities, resulting in a total of 12 districts for our survey. In each of the selected districts, we recruited patients with hypertension or diabetes from primary health care providers. Inclusion criteria included: (1) adults aged 18 years and above. (2) ability to express themselves clearly. (3) taking medicines for a long time (more than 3 months) due to chronic diseases. A total of 1,260 invitations were sent to patients with chronic diseases through primary care in all sample districts, resulting in 1205 patients agreeing to participate, a response rate of 95.6%. All of participants completed the questionnaires, 9 of which were excluded due to incomplete information. Effective responsive rate was 99.3%. All participants in this survey were required to complete an informed consent form or provide verbal consent to participate in the survey.

Demographic data

Demographic and disease-related data, including age, gender, education, domicile, residence status, job type, medical insurance, disease course, exercise, and drink, were collected through our questionnaire. In addition, according to Age-Based Grou** Criteria of World Health Organization and the average life expectancy in China at the end of 2019 [18, 19], we categorized participants into three age groups: adults (< 65 years), young-old (65–75 years), and oldest-old (≥ 75 years). We categorized patients’ residence status as living alone or not living alone according to whether they live with family.

Measurement of variables

We used the Control Preference Scale-Post (CPSpost) [20, 21], a modified version of Control Preference Scale (CPS), to assess the type of decision-making models of patients with chronic diseases, that is, the actual control of doctors and patients over medication decisions, which was perceived by patients. Previous studies indicated that CPSpost is a valid and reliable scale to measure the participation of patients in medical decision-making [21, 22]. A total of five entries are included as follows: (1) I made my medication decision alone; (2) I made my medication decision alone considering what my doctor said; (3) I shared the medication decision with my doctor; (4) My doctor decided considering my preferences; (5) My doctor made the medication decision. (1) and (2) were categorized as PDM (a process in which the patient is the initiative role in the decision-making), (3) was categorized as SDM, (4) and (5) were categorized as DDM (a process in which the doctor is the initiative role in the decision-making) [23]. We asked patients to answer based on two experiences as follows: (1) communication with doctors when prescribed medications for the first time; (2) communication with doctors during medication adjustments over the past three months.

Medication knowledge was evaluated through a questionnaire adapted from the study by McPherson et al., a total of seven entries and codes were shown in Table 1. According to McPherson’s classification method, we used the median score as the threshold to separate medication knowledge into high and low score groups of medication knowledge [24].

Table 1 Medication knowledge questionnaire
Full size table

Medication compliance was examined using the 4-item Morisky medication adherence scale (MMAS-4) [25], which was widely used to measure the medication compliance of patients with chronic diseases and has presented favorable among Chinese patients [26]. The detailed entries of the scale were as follows: (1) Do you ever forget to take your medicine? (2) Are you careless at times about taking your medicine? (3) When you feel better do you sometimes stop taking your medicine? (4) Sometimes if you feel worse when you take the medicine, do you stop taking it? For each item, we assigned the answer of “Yes” as 0 point, and assigned the answer of “No” as 1 point. We divided medication compliance into high and low score groups on medication compliance according to the distribution of total score.

Depression symptom was measured by Short Version of Center for Epidemiological Studies Depression Scale (CESD10), which met strict clinical requirement [27].

Statistical analysis

We used Pearson’s χ2 test to conduct descriptive analysis of demographic characteristics and other variables in different decision-making model groups.

Random forest (RF) is a machine learning method for noise immunity, prevention of over­fitting, and independence from co-linearity, which showed a preference for important predictor variables by Gini coefficient, and applied to any significance tests and variable selection [28]. We used the RF method for two main reasons. Firstly, when compared with other variable selection models, RF is a machine learning method that covers the impact of each predictor variable individually as well as in multivariate interactions with other predictor variables and thus work towards the global optimality of the variable selection [29]. Secondly, RF provides relative importance among variables, which is of great value in targeting interventions. In this study, we used Mean Decrease Gini (MDG) and out-of-bag (OOB) curve to select variables, which was proposed by Hong Han et al. [29]. Gini coefficient is an indicator reflecting to the degree of inconsistency in the sample categories on the node, the lower of Gini coefficient, the better results of classification [30]. MDG refers to the total decrease of Gini from splitting on the variable averaged over all trees, which is used to indicate the importance of the predictor variable to the response variable [29]. OOB error rate is used to estimate the prediction error of current model by using the set of remaining samples which are not included in current tree [29]. We firstly referred to MDG to rank the importance of predictor variables, then selected the most appropriate number of variables to be included in the multivariate logistic regression model according to OOB curve. The multivariate logistic regression with P value, OR, and 95% CIs was used for analyzing the predictor variables’ effect direction and relative hazard.

Moreover, due to the proportion of decision-making models is unbalanced (62 PDMs, 452 SDMs and 741 DDMs). The imbalance of categories will affect the classification effect of RF —— the classification result tends to favor the majority category. Therefore, we used the synthetic minority sample oversampling method (SMOTE) to balance the data. The SMOTE method is a data preprocessing technique applied to imbalance problems proposed by Chawla et al. [31], which uses the K- nearest neighbors and linear interpolation to add minority class samples to balance the class distribution [32]. In R’s smotefamily package, we set the K parameter to 3 and dup_size to 6, which means the minority class will generate 6 times as many new samples based on 3 original samples from random nearest neighbors. Finally, we obtained 372 new PDMs, a total of 432 PDMs included in RF.

R (version 4.0.3, R Project for Statistical Computing) and SPSS (version 24.0) were used for all statistical analyses in this study.

Results

Among 1196 participants, 57.94% participants were female, 48.33% resided in the urban area, and 62.63% were manual workers. The average age of participants was 68.55 years old (ranging from 26 to 92 years). The majority of patients used DDM (57.02%), while some patients used SDM (37.79%), and a smaller percentage opted for PDM (5.18%). Other detailed information was shown in Table 2.

Table 2 Characteristics of the study population
Full size table

Importance ranking of the independent variables

Ten independent variables with a p-value less than 0.05 in univariate analysis were included in RF analysis, with ntree as 500. Figure 1 showed the visualization results of the importance ranking of 10 variables. According to the results of MDG, the top 5 important variables were age, education, exercise, disease course, and medication knowledge.

Fig. 1
figure 1

The importance of factors influencing decision-making on medication

Full size image

Figure 2 showed that OBB error rate was lowest when model contained 5 variables. The top 5 variables in order of importance ranking were: age, education, exercise, disease course, and medication knowledge.

Fig. 2
figure 2

Relationship of OOB error rate with number of variables

Full size image

According to the importance ranking and OOB curve of RF, 5 independent variables (age, education, exercise, disease course, and medication knowledge) were included in multivariate logistic regression.

Influencing factors of decision-making model

We chose the SDM group as a reference group in multivariate logistic regression analysis since SDM was considered to be a hallmark of patient-centered care and more advocated during the clinical encounter compared with other two decision-making models.

Table 3 showed the significance of factors influencing chronic patients’ participation in medication decision-making. The patients with lower medication knowledge (OR = 2.737, P < 0.05) were more likely to use PDM than SDM.

Table 3 Multivariate logistic regression of factors associated with decision-making on medication
Full size table

When compared with DDM, the patients under 65 years (OR = 0.636, P < 0.005) and disease course under 10 years (OR = 0.750, P < 0.005) were more likely to participate in the SDM during the medication decision-making process. By contrast, patients with infrequent exercise (OR = 1.443, P < 0.05), lower educational levels (OR=1.536, P<0.05) and poor medication knowledge (OR = 1.446, P < 0.05), were more likely to use DDM.

Discussion

In this study, 57.02% of patients used DDM during their decision-making process, which was lower than similar research conducted in other Asian countries, such as the United Arab Emirates and Japan [33, 34]. It may be related to the success of the National Essential Public Health Service launched by the Chinese government in the last decade, which enhanced the family doctor signing rate and residents’ health literacy [62], while previous studies reported that there was limited effectiveness of education in SDM process due to pressurized healthcare environment and inadequate capacities of medical staff [13, 63]. Therefore, better preparation for decision-making, such as providing patient decision aids (PDAs), was more advocated [13]. Although PDAs have attracted the attention of researchers since the 1990s [64]. Few PDAs were designed for chronic disease patients in mainland China at present, esecially for their medication decision-making [65]. Therefore, develo** PDAs for medication decision-making in the context of Chinese cultural background and healthcare system would be a meaningful research direction in the future.

Strengths and limitations

In this study, we used a combination of RF and logistic regression model to find out the key factors associated with the participation in medication decision-making of patients with chronic diseases. This was a special feature compared with other SDM related research. The data were collected from 12 districts of 4 cities in the Hubei Province, China, which are representative of all patients with chronic disease in entire Hubei Province. However, there are still several limitations in this study. Firstly, this study only focused on the impact factors of participation in decision-making from the patients’ perspective, not paying much attention to the factors from healthcare providers. Secondly, this study recruited patients voluntarily. The patients who weren’t willing to participate were not surveyed, which may generate some data bias. Finally, logistic regression did not reported significance in some PDM results, which may be attributed to the small sample size of patients using PDM (only 62 cases, 5.18% of the total sample). It is important to expand the PDM sample in future studies.

Conclusion

According to the findings in this study, the key factors associated with SDM were age, education, exercise, disease course, and medication knowledge. Based on the results, several corresponding interventions could be taken to improve patients' participation in medication decision-making. Firstly, doctors should pay more attention to elderly patients with lower education levels, and encourage them to participate in SDM. Secondly, health education should focus on transforming patients’ traditional perceptions and behaviors to enhance their awareness of participation in SDM. Finally, development and application of PDAs to improve patients' medication knowledge and promote them to participate in SDM will be an important topic in further research and clinical practice.