A Research Based on Online Medical Platform: The Influence of Strong and Weak Ties Information on Patients’ Consultation Behavior

As an indispensable part of contemporary medical services, Internet-based medical platforms can provide patients with a full range of multi-disciplinary and multi-modal treatment services. Along with the emergence of many healthcare influencers and the increasing connection between online and offline consultations, the operation of individual physicians and their teams on Internet-based medical platforms has started to attract a lot of attention. The purpose of this paper is to, based on an Internet platform, study how the information on physicians’ homepages influences patients’ consultation behavior, so as to provide suggestions for the construction of physicians’ personal websites. We distinguish variables into strong- and weak-ties types, dependent on whether deep social interactions between physicians and patients have happened. If there exist further social interactions, we define the variable as the “strong ties” type, otherwise, “weak ties”. The patients’ consultation behavior will be expressed as the volume of online consultation, i.e., the number of patients. We obtained the strong and weak ties information of each physician based on EWM (entropy weight method), so as to establish a regression model with explained variable, i.e., the number of patients, and three explanatory variables, i.e., the strong and weak ties information, and their interaction term. The estimation results verified our hypotheses and proved to be robust. It showed that both strong and weak ties information can positively influence patients’ consultation behavior, and the influence of weak ties information is greater. Regarding the positive influence of strong and weak ties, we found a trade off effect between them. Based on the results, we finalize with some suggestions on how to improve a physician’s online medical consultation volume.


Introduction
Online healthcare consultation has become an essential part of healthcare system. Online medical platforms provide patients with a channel that allows them to make an appointment, learn about a physician, understand their severity of illness, and ask for advice on the Internet without having to leave home [1][2][3].
Researchers focus on different kinds of information in the online health community to investigate how factors impact patients' consultation behavior. The Information of physicians contains the self-disclosed information, online image, etc. [1,2,4,5]. In addition, consumer value was co-created by the online medical community [6].
The current physician-patient relationship in China is experiencing dilemmas, which are partly due to the information asymmetry [7][8][9] between physicians and patients. A physician's personal homepage can provide a channel for patients to learn about physicians and diseases, thus providing a service for patients to choose a physician based on detailed information [10,11], which, to some extent, alleviates information asymmetry and also exerts an influence on patients' consultation behavior [12,13]. The literature has divided behavior in the online medical field. Furthermore, the results, to some extent, can provide the support of theoretical suggestions to solve physician-patient conflicts caused by the information gap between physicians and patients in the Internet medical field.
Our study will explore the answers to the following questions: (1) How do strong and weak ties information affect patients' consultation behavior through the online medical platform? (2) Which has more influence on patients' consultation behavior? Strong ties information or weak ties information? (3) Do strong and weak ties exert a trade-off effect on each other's influence on patients' consultation behavior? For example, does the enhancement of weak ties information reduce the positive effect of strong ties information on the patients' consultation behavior? (4) The influence of non-social ties information, i.e., some nature of the physician himself, on the patients' consultation behavior.

Data Sources
We used Python to crawl 52,645 physicians' homepages on 18 January 2022 and 25 January 2022 from the well-known Chinese Online Medical Platform A for model building and robustness test. From the perspective of horizontal physicians' homepages, each datum contains numerical and textual information that users can see when they visit physicians' homepages. From the platform perspective, the 52,645 data we crawled account for 22% of the total number of registered physicians, which is a very large sample. Such a rich and large volume of data will help us to more fully explore the information behind the data and explore the impact of physicians' homepage information on patients' consultation behavior.

Variables
We select "Number of checked patients (NCP)", "Number of comments after consultation (NCC)", "Number of thank-you letters (NTL)", "Number of gifts (NOG)", and "Number of online patients (NOP)" to form strong ties. The variables such as "Comprehensive recommendation score (CRS)", "Number of articles on Health subscription (NAH)", and "Number of articles reads (NAR)" are used to form weak ties. These variables will be processed later to form three independent variables: WeakTies, StrongTies, and the interaction term WeakStrongTies, which will be the real independent variables needed for the models.
In this paper, we choose the number of patients who have consulted physicians as the dependent variable. This is the original collected variable that is automatically displayed on the physician's homepage, and can indicate the number of online consultation that the physician has received. We use this variable to measure the patients' consultation behavior. A higher number of patients indicates a more active patient consultation.
We also add dummy variables to the model. The title, education title, and outpatient information are set as dummy variables. We coded title dummy from 1 to 4, e.g., title_dummy1, coded education title dummy from 1 to 6, edutitle_dummy1. In order to measure the information of the highest level of outpatient consultation type that a physician can provide offline, we coded outpatient dummy from 1 to 8, e.g., op_dummy1. Table 1 shows the definitions and descriptive statistics of all original variables.

Hypothesis Strong Ties Models and Hypothesis
To explore how information on physicians' homepage impact patients' consultation behavior, we distinguish the initial information into strong ties information and weak ties information based on whether deep social interactions between physicians and patients have happened. If there exist deep social interactions, we define these kinds of initial variables as "strong ties" variables; otherwise, we define those initial variables as "weak ties" variables. In online medical platforms, patients and physicians will have close interactions, including but not limited to offline consultation, evaluation of physician's service, following physicians, expressing gratitude through thank-you letters and gifts, and so on. Therefore, this paper selects the patient-generated information data, i.e., "Number of checked patients (NCP)", "Number of comments after consultation (NCC)", "Number of thank-you letters (NTL)", "Number of gifts (NOG)", and "Number of followers (NOF)", generated by the indepth interaction between the physician and the patient to form the "strong ties" behavior indicator. Thus, the following hypotheses are proposed: Hypothesis 1. Strong ties have a positive effect on the patients' consultation behavior [33].

Weak Ties
System-generated information exists on the physician's homepage, such as the number of reads of articles published on the physician's website health subscription (NAR). Although some of these data, such as the comprehensive recommendation score rated by Online Medical Platform A, are jointly calculated by the algorithm based on some patientgenerated information and system-generated information on the physician's homepage. However, in general, this part of the data is mostly determined by the system, and the interaction between patients and physicians cannot exert too much influence on it. So, we use the explicit website data, i.e., "Comprehensive recommendation score (CRS)", "Number of articles on Health subscription (NAH)", and "Number of articles reads (NAR)", to form "weak ties" behavior indicators. Thus, the following hypotheses are proposed: Hypothesis 2. Weak ties have a positive effect on the patients' consultation behavior.

Interaction of Weak Ties and Strong Ties
System-generated information and patient-generated information will have an effect on each other's influence on the patients' consultation behavior, so we plan to investigate the interaction of strong and weak ties. Thus, we propose the following hypothesis: Hypothesis 3. The enhancement of weak ties information will reduce the positive effect of strong ties information on the patients' consultation behavior [34]. However, in general, this part of the data is mostly determined by the system, and the interaction between patients and physicians cannot exert too much influence on it. So, we use the explicit website data, i.e., "Comprehensive recommendation score (CRS)", "Number of articles on Health subscription (NAH)", and "Number of articles reads (NAR)", to form "weak ties" behavior indicators. Thus, the following hypotheses are proposed: Hypothesis 2. Weak ties have a positive effect on the patients' consultation behavior.

Interaction of Weak Ties and Strong Ties
System-generated information and patient-generated information will have an effect on each other's influence on the patients' consultation behavior, so we plan to investigate the interaction of strong and weak ties. Thus, we propose the following hypothesis: The enhancement of weak ties information will reduce the positive effect of strong ties information on the patients' consultation behavior [34].

Data Pre-Processing
When the total number of visits or total number of patients of a physician is too small, there will be a shortage of multiple data for this physician, so his or her data are not referenceable. Therefore, physicians with total visits less than 1000 were considered as abnormal data and were deleted. Finally, 42,319 pieces of data were retained.
After data cleaning, Python's sklearn.preprocessing package was used to scale the values and take the logarithm of the factor patients, thus completing the normalization of the data.

Strong and Weak Ties Model
In the process of constructing the strong and weak ties model, considering that the information performance of the original variables is fairly objective, we did not choose the subjective weighting method. Instead, we chose the objective weighting entropy method

Data Pre-Processing
When the total number of visits or total number of patients of a physician is too small, there will be a shortage of multiple data for this physician, so his or her data are not referenceable. Therefore, physicians with total visits less than 1000 were considered as abnormal data and were deleted. Finally, 42,319 pieces of data were retained.
After data cleaning, Python's sklearn.preprocessing package was used to scale the values and take the logarithm of the factor patients, thus completing the normalization of the data.

Strong and Weak Ties Model
In the process of constructing the strong and weak ties model, considering that the information performance of the original variables is fairly objective, we did not choose the subjective weighting method. Instead, we chose the objective weighting entropy method to assign weights to the original indicators and complete the classification calculation based on the coefficients, so as to build the required strong and weak ties model: (1) Strong Ties Model: (2) Weak Ties Model: (3) Interaction Term of Weak Ties and Strong Ties: We used Python to calculate the entropy method assignment coefficients and bring them into the Formulas (1) and (2), respectively, to obtain the following models: (4) Strong Ties Model: Healthcare 2022, 10, 977 6 of 15 (5) Weak Ties Model: Finally, we obtained the data of three variables, StrongTies i , WeakTies i , and StrongWeakTies i . We used these three as partial independent variables and participated in the construction of the regression model. Table 2 shows the definitions of the Variables and summary statistics, and Table 3 presents the correlations of the Variables. The variables "op-dummy*" are the dummy variables generated by the variable "outpatient", which represents the highest level of outpatient consultation type that a physician can provide offline. As shown in Table 2, e.g., "op_dummy1" represents the number of physicians whose highest level of outpatient consultation type is "VIP5-VIP outpatient".

Regression Model
We construct models 1-5 separately using ordinary least squares (OLS) regression.

Model 1
Since the number of consultations on online medical platforms represents the attractiveness of physicians to patients, we constructed five models to test how two types of information, weak ties and strong ties information, impact patients' consultation behavior. Model 1, which is our baseline model, is configured as follows: In this model, i = 1, 2, . . . , 5 presents physicians' id numbers. The dependent variable "Patients i " is the total number of patients who have consulted physicians. Variable "StrongTies i " is generated by Equation (1), and variable "WeakTies i " is generated by Equation (2).

Model 2
Considering that some confounding factors may have an influence on the relationship between social ties information and patients' consultation behavior, we chose physicians' title information as a dummy variable to join the model:

Model 3
The level of physicians' education title may influence the decision of patients' consultation behavior, so we added the education title as a dummy variable in model 3:

Model 4
Information of outpatient, which is the highest level of outpatient consultation type that a physician can provide offline, may also influence the decision on patients' consultation behavior, so we used outpatient information as another dummy variable in Model 4: 2.5.5. Model 5 To explore the nature of the interaction between the influences of education title and strong ties, we introduced six interaction terms on the basis of Model 4:

Results
The results are shown in Table 4. From experiments 1-5, it is seen that the coefficients for WeakTies and StrongTies are positive and significant, and the coefficients for Strong-WeakTies are negative and significant. Therefore, hypotheses 1-3 are accepted. In addition, the coefficients for weak ties are on average 9.8 times stronger than the coefficients for strong ties.  Robust t-statistics in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.
It can be perceived that in Model 1, WeakTies has a positive and significant coefficient, suggesting that weak ties information positively affects patients' consultation behavior (α 1 = 1528.062, p < 0.01). StrongTies also has a positive and significant coefficient, suggesting that strong ties information positively affects patients' consultation behavior (α 2 = 155.668, p < 0.01). The coefficient of StrongWeakTies is negative and significant (α 3 = −25, 141.955, p < 0.01), and is larger than α 1 and α 2 . This is because the coefficients of WeakTies and StrongTies are between (0, 1), the value is smaller after multiplication, and the coefficient is greater in the regression. This result shows that in the online medical platform, the enhancement of weak ties information will reduce the positive effect of strong ties on the impact of physicians' online consultations and vice versa.
In Models 2 to 4, we add the dummy variables of title, education title, and outpatient type step by step. In addition, we add the cross term in Model 5. We also test R 2 , AIC, and BIC. R 2 becomes bigger from 0.474 to 0.484, and AIC performs smaller from 173,591.7 to 172,810.8 in Models 1 to 5. BIC becomes smaller from 173,626.3 to 173,016.3 in Model 1 to 4 but becomes bigger in Model 5 (173,053). Considering these three tests, we think Model 5 performs the best, which indicates that education title information is useful to patients' consultation behavior.

Robustness Test
We adopted the method of replacing samples for robustness testing and selected 52,645 homepage information of the same group of physicians on 25 January 2022 to construct the same model for validation.
As shown in the Table 5, all the conclusions are consistent with Models 1-5 in the original experiment. Only in Models 3-5 was the coefficient of education title dummy of associate researcher performs significant, which is the same as other education title and does not affect the conclusion of the original variable analysis.  Robust t-statistics in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

Discussion
Through the above experiments, we can see that both strong and weak ties information has a positive effect on patients' decision of whether to choose the physician for consultation [14]. The information displayed on a physician's homepage on the online medical website helps patients to judge the physician's professional competence and service attitude and thus make a physician consultant decision.
Furthermore, we found that weak ties information is, on average, 9.8 times more influential than strong ties information in comparison to the influence on patients' consultation behavior. Therefore, we can conclude that strong ties information does not have a strong positive relationship with patients' physician consultation behavior. Instead, weak ties information has a greater positive impact on patients' consultation behavior. That is, it is not the case that the stronger the interaction between the physician and the patient, the greater the positive effect of the generated information on the patient's choice of the physician for consultation. In other words, the more intuitive information, that is, weak tie information, displayed on the physician's homepage without deep interaction between the physician and the patient, the stronger the positive effect on patients' consultation behavior. This also validates the findings of Granovetter (1973) [25], whose study concluded that weak interpersonal relationships are more important in the process of information dissemination than strong relationships. In addition, it is clear that if we change the initial variables, their influence on dependent variables will also be changed, which can be seen from the coefficients. However, the correlativity between weak and strong ties information will not change. For example, the coefficient of "WeakTies" is always bigger than the coefficient of "StrongTies".
On the other hand, this also shows that it is necessary for physicians to take some actions, such as proactively generating objective data, adding more expertise, etc., to maintain their homepage, so as to facilitate patients to understand their state of disease and judge the physician's professionalism. For example, patients can obtain a deeper understanding of a physician's knowledge of a certain disease from articles published on Health subscription, which is more effective than the information about the physician's interaction with the patient.
We also found that even though both strong and weak ties information have a positive effect on patients' consultation behavior, the enhancement of one side's information will reduce the positive effect of the other side's information on the consultation service. Strong ties information can certainly show the physician's high quality consultation service and patient's recognition, but combined with the conclusion that weak ties information is on average 9.8 times more influential than strong ties information, we suggest that physicians need to pay more attention to the role of weak ties information. Strong ties information is more concerned with the good results brought by the quality of consultation service, while weak ties information should be regarded as the focus of personal brand maintenance, so as to increase the number of consultations most efficiently.
However, in the exploration of dummy variables, we found that information such as title, education title, and outpatient type can impact the choice of patients slightly. In addition, the outpatient type and title information are more important than the education title. They all have positive impacts on patients' consultation behavior.
In this paper, we innovatively study the patients' consultation behavior from the perspective of social ties and explore the influence of strong and weak ties information on it. However, our study still has certain limitations. First, our data sources are limited, and we only crawled the data of Online Medical Platform A, a well-known Chinese online healthcare platform. However, there are important differences across the different media [30], and big data for health services are now becoming very popular [35]. As a result, whether the findings are applicable to all online healthcare platforms needs to be discussed in further studies. Second, the dummy variables we included in the model were all considered to show the objective professionalism of physicians. We explore the influence of title, education title, and the highest level of offline outpatient types on patients' consultation behavior, but the exploration of other information about physicians is insufficient. Third, we may consider the effect of physicians' group performance in the future [36].
In summary, our research provides support for further exploration of the role of strong and weak ties information on users in the online healthcare field. Future research could further explore how strong and weak ties affect other patient behaviors. It could also further mine textual information to explore the impact of online social text information on patient behavior.

Conclusions
The following conclusions and suggestions can be drawn from this paper: Both strong and weak ties have a positive effect on patients' consultation behavior. However, the enhancement of informativeness on one side will reduce the positive effect of information on the other side on the patients' consultation behavior.
In comparison of the impact on the patients' consultation behavior, weak ties information was on average 9.8 times more influential than strong ties information. Weak ties variables have a more positive effect on patients' consultation decision.
Based on the first and second conclusions, this paper suggests that physicians should pay more attention to improving the quality of weak ties information, to efficiently attract more patients and increase the number of consultations.
Information such as the physician 's title, education level and the highest type of outpatient physicians can offer offline, which can represent the physician 's professional competence, can have a positive impact on patients' physician consultation decision. The patients' consultation behavior can reflect patients' recognition of the physician. Therefore, this paper suggests that physicians should strive to improve their professional competence and obtain a high professional title and to increase patients' recognition and trust.

Data Availability Statement:
The image data used to support the findings of this study are available from the corresponding author upon request. The platform where the information is collected has been named by a particular name: Online Medical Platform A. All data have been desensitized, and there is no content that has privacy issues for any of the related parties.

Conflicts of Interest:
The authors declare no conflict of interest.