Symptom network topological features predict the effectiveness of herbal treatment for pediatric cough

Abstract

Pediatric cough is a heterogeneous condition in terms of symptoms and the underlying disease mechanisms. Symptom phenotypes hold complicated interactions between each other to form an intricate network structure. This study aims to investigate whether the network structure of pediatric cough symptoms is associated with the prognosis and outcome of patients. A total of 384 cases were derived from the electronic medical records of a highly experienced traditional Chinese medicine (TCM) physician. The data were divided into two groups according to the therapeutic effect, namely, an invalid group (group A with 40 cases of poor efficacy) and a valid group (group B with 344 cases of good efficacy). Several well-established analysis methods, namely, statistical test, correlation analysis, and complex network analysis, were used to analyze the data. This study reports that symptom networks of patients with pediatric cough are related to the effectiveness of treatment: a dense network of symptoms is associated with great difficulty in treatment. Interventions with the most different symptoms in the symptom network may have improved therapeutic effects.

Keywords

pediatric cough / complex network / symptoms / traditional Chinese medicine / electronic medical records

Introduction

Cough is a common symptom of pediatric respiratory system disease, which is also considered the disease condition in the category of pediatric pulmonary disease of traditional Chinese medicine (TCM). It is a protective reflex, which is a component of normal respiratory physiology that enhances mucociliary function and clears excessive secretions and airway debris from the respiratory tract and is a very common symptom of respiratory disease [¹,²]. Most episodes of cough in children are acute (less than 2 weeks) and secondary to a lower respiratory infection. A cough that persists beyond 2 weeks can be classified as prolonged acute (2–4 weeks) or chronic (more than 4 weeks) [³,⁴].

At present, most TCM clinical studies on pediatric cough focus on evaluating or confirming the effectiveness of different treatment principles and prescriptions. For example, Feng et al. explored the empirical knowledge of a highly experienced physician in treating pediatric cough with classical prescription [5]. Although pediatric cough has been investigated in various fields (e.g., etiology, symptom, and therapy), its incidence remains high. Highly experienced TCM physicians may also encounter some patients who have poor efficacy. Current studies investigate the relationship between the independent effect of symptoms and the diagnosis and treatment of symptoms. The diversity of the co-occurred symptoms of patients may be the cause for different treatment effects. In TCM clinical settings, the accompanying symptoms, such as nasal congestion and expectoration, of pediatric cough patients have significant contributions to syndrome (i.e., Zheng) differentiations, which involve complicated interactions among different symptoms [⁶].

Network medicine, which aims to gain an understanding of human disease from the perspectives of systems biology and complex networks, is a rapidly developing field that has become increasingly important for identifying novel disease mechanisms and predicting drug efficacy [⁷,⁸]. For example, one recent study proposed a symptom network approach to improve the understanding of mental disorders; in this approach, disorders are conceptualized as complex dynamic systems of interacting symptoms [⁹]. According to a related study on mental disorders [¹⁰], a strongly connected symptom network may be associated with less positive prospects. If this situation is held true for pediatric cough, then we should expect symptoms to be strongly connected within the pediatric cough groups that have worse efficacy.

Here, we investigate a symptom network of pediatric cough and evaluate its association with clinical effectiveness for treatment. In this study, similar to the work on mental disorders [¹⁰], overall network connectivity was compared using the recently developed Network Comparison Test (NCT) [¹¹]. Local connectivity of individual symptoms in the network was compared using four centrality measures (node betweenness, closeness, strength, and eigenvector centrality) [¹²].

Materials and methods

Study sample

The data set was selected from the electronic medical records of a TCM physician (co-author of this work) from November 2006 to June 2014 in the Hubei Provincial Hospital of Traditional Chinese Medicine and the Guangzhou Clifford Hospital, with permission. A total of 496 patients with 1640 cases were screened for the diagnosis of pediatric cough.

Using the platform for handling medical records [¹³] and checking the original medical records, we selected 254 patients, 384 cases (patient-visits) for this study, on the basis of the following inclusion criteria: (1) the diagnosis was cough, (2) the case had a therapeutic effect, and (3) Chinese herbal treatments were used in every case. The exclusion criteria were (1) the case only had one medical record and (2) these cases, diagnosed cough, were actually cold [¹⁴], asthma [¹⁵], and pneumonia [¹⁶]. To evaluate the effectiveness of different cases, we considered the change in symptoms between the sequential consultations as the main indicator. For example, cough was the main symptom of pediatric cough. If “cough reduction,” “occasional cough,” or “no cough” appeared in the subsequent visit, it was considered effective. By contrast, “cough aggravation” or “no change of cough occurred” indicated ineffective. The evaluation of efficacy was divided into four categories, namely, recovery, improvement, no change, and deterioration. Recovery and improvement were categorized as valid, whereas no change and deterioration were invalid. This study was a retrospective case research; thus, the therapeutic effect was assessed on the basis of the symptoms change. After evaluation, we obtained 384 cases (patient-visits) that met the above-mentioned criteria; 344 cases were considered valid, and the remaining cases were considered invalid. The number of patients and cases before and after evaluation is shown in Table 1, and the interval time between one patient-visit and its subsequent patient-visit is shown in Table 2.

In our study, if one symptom occurred in a case, then the score of this symptom is 1, and the opposite is 0. We used the Pearson’s correlation coefficient to measure the correlation between two symptoms. The coefficient is the covariance of the two variables divided by the product of their standard deviations. The formula of the correlation between symptom X and symptom Y is

(1)\rho (X,Y)={\displaystyle \frac{cov(X,Y)}{{\sigma }_X{\sigma }_Y}},

where cov(X,Y) is the covariance, {\sigma }_X is the standard deviation of X, and {\sigma }_Y is the standard deviation of Y. The Pearson’s correlation coefficient ranged from −1 to 1 [¹⁷].

However, {\sigma }_X was 0 if symptom X occurred in all cases or not in a group, which indicated that symptom X is a constant. For example, if all cases had the symptom of cough, then no correlation was observed between cough and other symptoms. Thus, we removed those symptoms from the follow-up analysis. A total of 52 symptoms remained. Table 3 lists the abbreviations and full names of the 52 symptoms. Groups A and B represent the invalid and valid groups, respectively. Supplementary Tables S1 and S2 show the deleted symptoms in both groups and full names of the deleted symptoms, respectively.

Statistical analysis

General differences

The Wilcoxon rank sum test, a non-parametric test that compares two paired groups [¹⁸], was used to test differences in each symptom between groups A and group B. The significance level for all analysis was 0.05. Statistical analysis was performed using R-package version 3.4.1 [¹²].

Network construction

The network structures were investigated separately for groups A and B using Pearson’s correlations among symptoms. In the network, the nodes represent symptoms and the edges represent the correlation between symptoms. The weight of edges indicates the absolute value of correlation between symptoms. The correlation networks were performed using Qgraph, which was implemented in R-package Qgraph [¹⁹].

Differences in overall connectivity

We used four overall connectivity measures, namely, global strength, average shortest path length, diameter, and density, to analyze the differences between the two symptom networks.

The global strength of the network is defined as the weighted sum of the absolute connections [²⁰]. The formula of global strength GS is

(2)GS={\displaystyle {\sum }_{i\ne j}{w}_{i,j},}

where w_{i,j} is the weight (the absolute value) between symptoms i and j.

The atverage shortest path length is the average number of steps along the shortest paths for all possible pairs of symptoms [²¹]. The formula of average shortest path length ASPL is

(3)ASPL={\displaystyle \frac{\mathrm{1}}{n\cdot (n-\mathrm{1})}}·{\displaystyle {\sum }_{i\ne j}{d}_{i,j},}

where n is the size of the graph, that is, the total number of symptoms in the network; and d_{i,j} denotes the shortest distance between symptoms i and j.

Diameter is the maximum shortest path length in the network, which is the greatest distance between any two symptoms [²²]. The formula of diameter DIA is

(4)DIA=\underset{i\ne j}{\max }{d}_{i,j}.

The density of a network is the ratio of the numbers of edges and possible edges [²²]. The formula of density DEN is

(5)DEN={\displaystyle \frac{m}{n\cdot (n-\mathrm{1})}},

where m is the total number of edges and n is the total number of nodes in the network.

Statistical assessment of the difference in overall connectivity between networks of both groups was performed using NCT, a two-tailed permutation test [23], in which the differences of the four measures between groups A and B were calculated 100 000 times for randomly regrouped individuals. This performance resulted in a distribution under the null hypothesis assuming that both groups were equal, which could be used to test the difference between two groups. The difference was considered significant at a threshold of 0.05.

Differences in local connectivity

We used four best local centrality measures, namely, node strength, closeness, betweenness, and eigenvector centrality, to reveal which symptoms are important in the corresponding symptom network.

Node strength is a direct metric to characterize the centrality in a network. The method of calculating strength is to sum up the edge weights of the adjacent edge for each node [²⁰]. When the strength is large, the degree centrality of this node is high, which indicates the node is more important than other nodes in the network. Here, the formula for the strength of symptom i is

(6)S_i={\displaystyle {\sum }_j{w}_{i,j}}.

Closeness centrality measures the number of steps required to access every other node from a given node. The closeness centrality of a node is defined by the inverse an average length of the shortest path from all the other nodes in the network. Thus, the more central a node is, the closer it is to all other nodes [²⁴]. The formula for the closeness of symptom i is

(7)C_i={\displaystyle \frac{\mathrm{1}}{{\displaystyle {\sum }_j{d}_{i,j}}}}.

Betweenness centrality quantifies the number of times a node acts as a bridge along the shortest path between two other nodes. A node with a high betweenness centrality will have strong control over the network [²²]. The formula for the betweenness of symptom i is

(8)B_i={\displaystyle {\sum }_{i\ne j\ne k}{\displaystyle \frac{{\sigma }_{jk}(i)}{{\sigma }_{jk}}},}

where {\sigma }_{jk} is total number of shortest paths from symptom j to symptom k and {\sigma }_{jk}(i) is the number of those paths that passes through symptom i.

However, the importance of a node also depends on the importance of its neighbors [²⁴,²⁵]. Eigenvector centrality measures the degree to which the node is connected to other central nodes; they correspond to the values of the first eigenvector of the graph adjacency matrix. This metric is proportional to the sum of the centrality of nodes connected to the focal node [²⁶]. Let M=(m_i,j) be the graph adjacency matrix, i.e., m_i,j is the absolute connection between symptom i and symptom j. The related eigenvector equation about matrix M is ME=λE, where λ is the unique largest eigenvalue and E is the corresponding eigenvector. The i^th component of the eigenvector E then gives the eigenvector centrality score of the symptom i in the symptom network.

Quantifying importance of symptoms

To quantify the importance of symptoms in two symptom networks, we calculated the effect size for differences in mean local centrality of symptoms between groups A and B. Symptoms with the highest effect size were considered the most distinctive. Here, data of groups A and B were resampled 1000 times with the bootstrap method [²⁷], which resulted in the distribution of all symptom local centrality measures in both groups. Cohen’s d [²⁷] was used to calculate the effect size of each symptom. Given that we considered node strength, closeness, betweenness, and eigenvector, the effect size of each symptom was the average of effect sizes in the four measures.

Results

General differences

The age and gender distributions of the 384 cases in groups A and B are shown in Tables 4 and 5, respectively. We performed a permutation test to analyze the differences in age and gender, and the significance level was 0.05. The P value of age was 0.44, and the P value of gender was 0.97; no significant difference was observed in age and gender.

Statistical analysis showed that the mean score value of 37 symptoms in group A was higher than that in group B (Table 6). A total of 50 symptoms (96.15%) had a P value>0.05, which indicated that the single symptom could not cause a significant difference in effectiveness. Only two symptoms, namely, thready pulse (thp) and pulmonary rales (pmr), had P values of less than 0.05.

Differences in overall connectivity

From the two symptom correlation networks, we found that the network of group A was more strongly connected than that of group B (Fig. 1). We counted the number of positive and negative correlation coefficients in two networks; the significance level was less than 0.05 (Table 7). We also selected the 20 highest absolute values of positive and negative correlations of the two networks (Table 8). The positive correlation value between symptoms of the invalid group was higher than that of the valid group. A large absolute value indicates a strong correlation. If the positive correlation is large, then the two symptoms will occur frequently. Otherwise, the occurrence will be less frequent. In the invalid group, seven pairs of symptoms had weights of 1. They were as follows: swollen lymph follicles of the throat (slft) and deep red tongue (dpg), throat secretions (ths) and barking cough (bcg), conjunctival congestion (cjc) and pharyngalgia (phg), yellow sputum (ysp) and fingerprint in life pass (flp), sleep disturbed (sld) and night crying (ncr), throat secretions (ths) and fine fingerprints (fif), and barking cough (bcg) and fine fingerprints (fif). However, the positive correlation values of group B were all less than 1.

As shown in Fig. 2, the average shortest path length, density, diameter, and global strength of group A were all higher than those of group B (Fig. 2). A significant difference in density, diameter, and global strength (P<0.05) was found between groups A and B (Table 9).

Differences in local connectivity

To investigate differences in local connectivity, we compared the networks of groups A and B on the basis of the four centrality measures (Fig. 3).

Similar patterns were found when node strength was analyzed (Fig. 3). By contrast, barking cough (bcg), fine fingerprints (fif), fingerprint in wind pass (fiw), loose stool (lst), nasal mucus (nam), thick coated tongue (tct), throat secretions (ths), and white-coated tongue (wct) had higher values in the invalid cases’ network. In terms of closeness (Fig. 3), all symptoms had a higher value in the invalid cases’ network. However, the closeness of the invalid cases’ network was close to 0. Regarding betweenness (Fig. 3), the value in the invalid cases’ network was high in the following symptoms: expectoration (exp), fingerprint in wind pass (fiw), lavender fingerprints (lvf), nasal mucus (nam), thick coated tongue (tct), and white-coated tongue (wct). Similar to closeness, the betweenness in the valid cases’ network was also close to 0. The eigenvector centrality had a different pattern in two networks (Fig. 3). The symptoms barking cough (bcg), fine fingerprints (fif), nasal mucus (nam), pink tongue (pkt), and throat secretions (ths) had a high value in the invalid cases’ network. In the valid cases’ network, symptoms that had a high value were fingerprint in wind pass (fiw), lavender fingerprints (lvf), slippery pulse (slp), and thready pulse (thp).

From all four centrality measures, two symptoms had high values of effect size, namely, expectoration (exp) and constipation (con), with effect sizes of 3.21 and 3.03, respectively (Table 10). Nasal discharge (nsd, effect size= 2.88), pulmonary breath sounds rough (pbs, effect size= 2.81), anorexia (ano, effect size= 2.61), antiadoncus (ant, effect size= 2.51), pharyngeal red (phr, effect size= 2.30), nasal obstruction (nao, effect size= 2.20), and sneezing (snz, effect size= 2.03) also had high effect size values (Table 10).

Discussion

TCM has a long history of viewing an individual or patient as a system with different statuses and has accumulated abundant clinical experience, thereby forming a comprehensive and unique medical system. Compared with western medicine, TCM usually focuses on functional clinical phenotype investigation, which constructs the evidence chain of TCM diagnosis and treatment [²⁸].

The physiologic characteristics of children include delicate and immature zang-fu organs, insufficient physique and qi, vigorous activity, and rapid growth and development. The main pathology that children manifest is that they are easily attacked by diseases. When they are sick, their condition tends to change quickly and becomes serious with improper treatment. However, children also have pure and clear zang qi that helps them to make a quick recovery. Therefore, conducting proper interventions on symptoms immediately is important [⁶].

In our study, we found two symptoms (pmr and thp) that were significantly different in different efficacy groups. Rhonchi is the sound produced by turbulence that occurs during inhalation or exhalation of the airflow due to the narrow or partial obstruction of the trachea or bronchi. When the airflow passes through the bronchus that contains thin secretions (exudate, mucus, pus, and blood), the sound that occurs when the liquid forms blisters and ruptures immediately is known as “moist crackles.” The two pulmonary rales reflect inflammation of trachea or bronchi seriously, which may result in significant differences between groups A and B. Thready pulse indicates physiological characteristics of insufficient physique, qi, and blood, which may be the reason for significant differences between the two groups. However, most symptoms in both groups did not show significant differences in effectiveness, which indicates that not a single symptom but the symptom interactions have an impact on efficacy [²⁹].

To our knowledge, this study is the first to use the network approach for identifying the symptom interactions as regards clinical effectiveness in TCM. Here, we find that the symptom network of pediatric cough patients with poor treatment effect (group A) is denser than that of patients with effective treatment (group B). In our study, the valid group with a sparse symptom network indicates weak symptom interaction; that is, symptoms in the valid group appear singly. Therefore, the symptom interaction of the valid group is lower than that of the invalid group, which may be an important cause for a good or poor curative effect.

Furthermore, we find that the symptoms, namely, expectoration (exp), constipation (con), nasal discharge (nsd), pulmonary breath sounds rough (pbs), anorexia (ano), antiadoncus (ant), pharyngeal red (phr), nasal obstruction (ano), and sneezing (snz), identified by local centrality measures, are more significant in group A than in B. All these symptoms with high effect sizes also represent the diversity of pediatric cough, and they also tend to be accompanied with various symptoms. This finding indicates that the complexity of the clinical pathogenesis may be the reason for the poor treatment effect. At the same time, these symptoms are particularly plausible for intervention. In the clinical process, those symptoms are accompanied by symptoms of pediatric cough, and doctors often intervene in them. Pulmonary breath sounding rough is a main point in the diagnosis of pediatric cough [⁶]. For instance, Professor Wang often uses common drug combinations to treat constipation and anorexia [³⁰]. Professor Nie estimates the nature of pediatric cough due to exopathy on the basis of the condition of laryngopharynx. In addition, he indicates that pharyngeal red represents that the nature of the cough is heat, which should be interfered [³¹]. Therefore, the information on local connectivity may guide clinical therapy. However, additional research is needed to confirm this hypothesis. For example, which centrality measure is clinically most relevant in identifying the importance of symptoms has yet to be established. Although a central symptom is likely to have an influence on other nodes, it may be an efficient target for intervention if associations with other symptoms are directed outward or are at least bidirectional [³²].

The key advantages and strengths of this study are as follows. Compared with the previous studies, we used the method of analyzing the network structure to investigate the relationship between symptoms and efficacy on pediatric cough. Our clinical data curated from Professor Ni and the previous studies [³³] showed that Professor Ni used the same therapeutic process to treat pediatric cough, which makes our work feasible by evaluating the effectiveness only from the clinical feature perspective. In addition, we performed statistical comparison based on NCT and effect size for differences that made our results reliable. However, our study has several potential limitations. First, TCM is environment sensitive, and our data are imperfect because of lacking fine details on their population distribution. In addition, this study was conducted on the basis of a relatively small volume of the retrospective data derived from a single TCM physician. Furthermore, no reliable method is available to determine what symptoms in both networks directly lead to good or poor outcomes, which has significant implications for TCM clinical practice. In the future work, we will incorporate additional cases from different physicians or attempt a well-designed prospective study to validate the findings. We also intend to develop a reasonable method to determine what symptoms in both networks directly lead to good or poor outcomes.

Conclusions

This study reports that symptom networks of patients with pediatric cough are related to the effectiveness of treatment: a dense baseline network of symptoms is associated with great difficulty in treatment. Additional symptom interaction may be an important reason for the difficulty in treating pediatric cough. This proof-of-principle concept on symptom network is a promising line of research and can provide feasible approaches for the TCM diagnosis and treatment of other diseases.

References

[1] Gibson PG, Chang AB, Glasgow NJ, Holmes PW, Katelaris P, Kemp AS, Landau LI, Mazzone S, Newcombe P, Van Asperen P, Vertigan AE; CICADA. CICADA: Cough in Children and Adults: Diagnosis and Assessment. Australian cough guidelines summary statement. Med J Aust 2010; 192(5): 265–271

[2] Chang AB, Landau LI, Van Asperen PP, Glasgow NJ, Robertson CF, Marchant JM, Mellis CM; Thoracic Society of Australia and New Zealand. Cough in children: definitions and clinical evaluation. Med J Aust 2006; 184(8): 398–403

[3] Chang AB, Glomb WB. Guidelines for evaluating chronic cough in pediatrics: ACCP evidence-based clinical practice guidelines. Chest 2006; 129(1 Suppl): 260S–283S

[4] Gilchrist FJ, Carroll WD. Assessing chronic cough in children. Paediatr Child Health 2016; 26(6): 273–275

[5] Feng G, Zheng H, Zheng QZ. Experience of ZHENG Qi-zhong in using Puji Xiaodu Decoction. China J Tradit Chin Med Pharm (Zhonghua Zhong Yi Yao Za Zhi) 2016; 31(7): 2615–2617 (in Chinese)

[6] Wang S, Qiao-Wong JM, Zhao X. Pediatrics in Chinese Medicine: PMPH-USA, 2012

[7] Barabási AL, Gulbahce N, Loscalzo J. Network medicine: a network-based approach to human disease. Nat Rev Genet 2011; 12(1): 56–68

[8] Borsboom D, Cramer AO. Network analysis: an integrative approach to the structure of psychopathology. Annu Rev Clin Psychol 2013; 9: 91–121

[9] Boschloo L, van Borkulo CD, Rhemtulla M, Keyes KM, Borsboom D, Schoevers RA. The network structure of symptoms of the diagnostic and statistical manual of mental disorders. PLoS One 2015; 10(9): e0137621

[10] van Borkulo C, Boschloo L, Borsboom D, Penninx BW, Waldorp LJ, Schoevers RA. Association of symptom network structure with the course of depression. JAMA Psychiatry 2015; 72(12): 1219–1226

[11] van Borkulo C D. NetworkComparisonTest: Statistical comparison of two networks based on three invariance measures. R package version 2.0.1. Retrieved from the website of CRAN-NetworkComparisonTest. 2016

[12] Team RDC. R: a language and environment for statistical computing. Computing 2013; 1: 12–21

[13] Zhou X, Chen S, Liu B, Zhang R, Wang Y, Li P, Guo Y, Zhang H, Gao Z, Yan X. Development of traditional Chinese medicine clinical data warehouse for medical knowledge discovery and decision support. Artif Intell Med 2010; 48(2-3): 139–152

[14] Lu Q. Consensus of experts on standardized diagnosis and treatment of common cold in Chinese children (2013). Chin J Prac Pediatr (Zhongguo Shi Yong Er Ke Za Zhi) 2013; 28(9): 680–681 (in Chinese)

[15] The Subspecialty Group of Respiratory Diseases, The Society of Pediatrics, Chinese Medical Association. Guidelines for the diagnosis and treatment of bronchial asthma in children (2016). Chin J Pediatr (Zhonghua Er Ke Za Zhi) 2016; 54(3): 168–173 (in Chinese)

[16] The Subspecialty Group of Respiratory Diseases, The Society of Pediatrics, Chinese Medical Association. Guidelines for the management of community-acquired pneumonia in children (2013). Chin J Pediatr (Zhonghua Er Ke Za Zhi) 2013; 51(10): 745–749 (in Chinese)

[17] Zhou X, Menche J, Barabási AL, Sharma A. Human symptoms-disease network. Nat Commun 2014; 5(1): 4212

[18] Mann HB, Whitney DR. On a test of whether one of two random variables is stochastically larger than the other. Ann Math Stat 1947; 18(1): 50–60

[19] Maass M, Gieffers J, Krause E, Engel PM, Bartels C, Solbach W. Qgraph: network visualizations of relationships in psychometric data. J Stat Softw 2012; 48(4): 1–18

[20] Barrat A, Barthélemy M, Pastor-Satorras R, Vespignani A. The architecture of complex weighted networks. Proc Natl Acad Sci USA 2004; 101(11): 3747–3752

[21] Watts DJ, Strogatz SH. Collective dynamics of ‘small-world’ networks. Nature 1998; 393(6684): 440

[22] Caldarelli G. Introduction to complex network. AIP Conf Proc 2003; 661: 17–23

[23] Good P. Permutation tests: a practical guide to resampling methods for testing hypotheses. Springer Science & Business Media, 2013

[24] Boccaletti S, Latora V, Moreno Y, Chavez M, Hwang DU. Complex networks: structure and dynamics. Phys Rep 2006; 424(4–5): 175–308

[25] Opsahl T, Agneessens F, Skvoretz J. Node centrality in weighted networks: generalizing degree and shortest paths. Soc Networks 2010; 32(3): 245–251

[26] Bonacich P. Power and centrality: a family of measures. Am J Sociol 1987; 92(5): 1170–1182

[27] Efron B, Tibshirani R J. An introduction to the bootstrap. CRC press, 1994

[28] Jiang M, Lu C, Zhang C, Yang J, Tan Y, Lu A, Chan K. Syndrome differentiation in modern research of traditional Chinese medicine. J Ethnopharmacol 2012; 140(3): 634–642

[29] Miaskowski C, Dodd M, Lee K. Symptom clusters: the new frontier in symptom management research. JNCI Monographs 2004; 2004(32): 17–21

[30] Zhang X, Wang X, Wei W, Zhang X. Analysis on regulation of Wang Xuefeng’s prescriptions on children children chronic cough by data mining technology. Liaoning J Traditi Chin Med (Liaoning Zhong Yi Za Zhi) 2017; 44(9): 1793–1795 (in Chinese)

[31] Zhang QX. Experience of professor NIE Hui-min in treating exogenous cough of children. China J Tradit Chin Med Pharm (Zhonghua Zhong Yi Yao Za Zhi) 2008; 23(4): 335–337 (in Chinese)

[32] Zhou XZ, Zhang RS, Shah J, Rajgor D, Wang YH, Pietrobon R, Liu BY, Chen J, Zhu JG, Gao RL. Patterns of herbal combination for the treatment of insomnia commonly employed by highly experienced Chinese medicine physicians. Chin J Integr Med 2011; 17(9): 655–662

[33] Liu JZ, Huang MX, Liu XY, Chen Y. The data mining analysis of Professor Ni Zhuying’s treatment of pediatric cough disease. World Sci Technol (Shi Jie Ke Xue Ji Shu—Zhong Yi Yao Xian Dai Hua Za Zhi) 2017; 19(9): 1527–1532 (in Chinese)

RIGHTS & PERMISSIONS

Higher Education Press and Springer-Verlag GmbH Germany, part of Springer Nature