We conducted a retrospective chart review of children who visited the Suwon Aijoa KM clinic located in Suwon, South Korea for treatment for height gain between January 2011 and April 2015. The Kyung Hee University Korean Medicine Hospital institutional review board (IRB) approved this retrospective observational study (KOMCIRB-160219-HR-009). The inclusion criteria were as follows: (1) age below 16 years; (2) final height percentile measurement recorded before September 30, 2015; (3) treatment duration of 2–12 months; and (4) presented at the Aijoa clinic for treatment for height gain. The exclusion criteria were as follows: (1) missing record of dependent variables; (2) history of GH deficiency; and (3) outliers, predefined by height percentile value less than the mean percentile by more than three times the SD.

The height and weight of the patients were recorded at baseline as well as at the end of herbal medicine treatment for height gain. Body mass index (BMI) was calculated using the electronic medical record (EMR) system. Data regarding sex, age, expression of secondary sex characteristics, and previous medical history including KM and GH treatment were obtained from the medical records of the patients. Midparental target height (MPH) was calculated automatically using the EMR system, according to the formula [Predicted height of a boy = (father’s height + mother’s height + 13) / 2; Predicted height of a girl = (father’s height + mother’s height − 13) / 2]16). Chronological age (CA) was also calculated automatically. All of the children who presented for treatment for height gain were referred to a external medicine clinic for the evaluation of bone age (BA) and Cobb’s angle using diagnostic X-ray equipment (REX-320R; Listem, Seoul, Korea)17). Bone age was evaluated based on the anteroposterior(AP) view X-ray images of the left hand and wrist, according to the Greulich-Pyle method18),19). Cobb’s angle and BA were evaluated by the same radiologist.

Height percentiles were calculated automatically using the EMR system according to the standard growth chart published by the Korean Pediatric Society in 2007 based on the results of a nationwide research20). Since the height percentiles were published on a monthly basis, the height percentiles of the children included in our study were re-calculated every month using the well-established modified LMS algorithm20),21). Percentiles were measured at baseline and at the end of herbal medicine therapy for height gain. The difference in height percentile before and after treatment was defined as the primary outcome variable. In secondary analysis, children with the upper and lower one-third values of differences in percentile before and after treatment were categorized into the responder and non-responder groups, respectively. Upon division of the tertile groups, in cases where the lowest percentile values of the upper group and the highest percentile values of the lower group exhibited similar values, all of the unique percentile values were included in the analysis.

Individualized herbal medicine in the form of water-extract decoction (60 mL; twice a day) was prescribed to each patient on the basis of pattern identification. The composition of the prescribed herbal medicine and treatment duration were also recorded. Treatment duration was defined as the number of prescribed herbal medicine packages divided by two, since every patient was instructed to ingest two packages of herbal medicine per day. Since treatment for rhinitis or obesity is often performed in conjunction with treatment for height gain, the usage of major medicinal herbs that treat obesity, such as Ephedra sinica, and rhinitis, such as Forsythiae fructus and Schizonepeta tenuifolia, was recorded in order to investigate the effects of these additional medicinal herbs on treatment response.

Herbal formula prescribed to each child was defined as the formula prescribed for more than 30% of the total treatment duration. The names of the herbal formulas have been presented in Table 5, along with the ‘methods of treatment (治法)’ according to traditional Chinese medicine theory. Description of the ‘methods of treatment’ followed the World Health Organization international standard terminologies on traditional medicine22).

Statistical analysis was performed using SPSS 18.0 for Windows (SPSS Inc., Chicago, IL, USA). While continuous variables among the baseline characteristics were expressed as mean values ± SD, minimum values, 25^th percentiles, 75^th percentiles, and maximum values, binary variables were expressed frequencies and percentages. The influence of each predictive variable on a dependent variable was evaluated by simple linear regression analysis. Predictive variables that exhibited p-values < 0.2 upon simple linear regression analysis were further evaluated by multiple linear regression analysis. The predictive variable for inclusion in the final model was selected using the stepwise regression method. Values of p < 0.05 were considered statistically significant. The R² and adjusted-R² values were also calculated. The optimal cutoff value for the prediction of response to treatment was determined by ROC curve analysis of the continuous variable that was included in the final model. The cutoff value was calculated according to the Youden index, which maximizes the sum of sensitivity and specificity23).

The feasibility of the regression model was verified by the following method. Autocorrelation was evaluated using the Durbin-Watson index; values between 1.8 and 2.2 indicated the absence of autocorrelation24),25). Multicollinearity was determined according to the variance inflation factor (VIF); values > 10 indicated multicollinearity26). In case of variables with VIF > 10, the variable with the highest VIF was excluded from multiple linear regression analysis. The normality of distribution of the residuals was evaluated by the Kolmogorov-Smirnov analysis, where, p-value > 0.1 indicated normality. The homoscedasticity of the residuals was qualitatively evaluated using graphs.

A total of 5,439 patients presented at the clinic, among whom, 93 visited for treatment for height gain and underwent measurement for height percentile at baseline and at the end of treatment. Of the 93 patients, only 61 (male, n = 25; mean age, 10.7 ± 2.4 years) satisfied the inclusion and exclusion criteria. Of the 61 patients, 3 had previously received KM therapy, while 1 had received GH therapy. The other basic characteristics of the patients are shown in Table 1.

In order to avoid multicollinearity, the representative variables were included in regression analysis. For example, height and weight were integrated into the BMI index. In simple linear regression, the p-values for sex, age, expression of secondary sex characteristics, MPH, BMI, CA minus BA, and administration of Forsythiae fructus and Schizonepeta tenuifolia were less than 0.2 (Table 2). Therefore, these eight variables were further analyzed by multiple regression analysis.

None of the variables exhibited autocorrelation and multicollinearity. Assumptions of normality and homoscedasticity of the residuals were satisfied. Upon multiple linear regression analysis by stepwise selection, patient age and administration of Forsythiae Fructus were revealed to be the only two variables included in the final model. The regression model was found to be statistically significant. The adjusted-R² value was 0.231 (Table 3). The regression equation of difference in percentile was as follows:

The differences in height percentile before and after treatment ranged from 3 to 14 in the responder group (n = 21) and −13 to −1 in the non-responder group (n = 24). While 7 of the children in the responder group exhibited a difference in height percentile of 3, 6 in the non-responder group exhibited a difference of −1. Therefore, although the numbers of lower and upper tertiles were both greater than 20, they were not equal. The average patient ages were 10.1 ± 2.0 years and 11.5 ± 2.2 years in the responder and non-responder groups, respectively. The difference in mean age between the two groups was statistically significant (p = 0.023; Figure 1). The area under curve of ROC curve (AUC) was 0.698 (95% Confidence Interval (CI), 0.537–0.860; p < 0.001). The optimal cutoff value of age, which exhibited maximum sensitivity (83.3%) and specificity (61.9%) for the prediction of response to herbal medicine treatment, was 9.75 years (Figure 2). This indicated that treatment response in children below 9.75 years of age was better compared to that in children older than 9.75 years.

Based on the cutoff value of age, we performed subgroup analysis in order to identify any age-related differences in treatment response. Analysis of covariance (ANCOVA) was performed for comparison of variables between the younger (age ≤ 9.75 years) and older (age > 9.75 years) groups of children. The baseline and post-treatment height percentiles were the covariate and dependent variables, respectively. Normality of distribution was confirmed by the Kolmogorov-Smirnov test (p > 0.05), and equality of variance was satisfied by Levene’s test (p > 0.05). The interaction term between the baseline percentile and age was not significant. Therefore, this interaction term was excluded in the final ANCOVA. Before adjustment for the baseline percentile, the height percentile was increased in the younger group (49.61–53.00), but decreased in the older group (45.47–44.58) of children. According to the results of ANCOVA after adjustment for the baseline percentile, the post-treatment height percentile was significantly higher in the younger group compared to that in the older group (p < 0.001; Table 4).

The dryness-moistening (潤燥劑) and tonifying and replenishing (補益劑) formulas are the most frequently used treatment methods for height gain in children. The most frequently prescribed herbal formula in the present study was modified Jaeumyangyeongtang (滋陰 養榮湯加減方), which was administered to 26 children. Detailed method of treatment and herbal formula for height gain in each child are shown in Table 5. Detailed composition of each herbal formula is presented in supplementary 1. Modified formula indicates “a formula derived from the original one with some addition or reduction of ingredients” 22), according to traditional East Asian medicine (TEAM) theory. To elaborate, depending on the symptoms of the children, some medicinal herbs were added or removed from the herbal formulas presented in supplementary 1. For example, if a child presented with rhinitis, Forsythiae fructus and Schizonepeta tenuifolia were added to the herbal formula based on the judgment of the physician.

The results of the present study showed that the age of initiation of herbal medicine treatment and administration of Forsythiae fructus are predictive factors for treatment response for height gain in children. The optimal cutoff value of age for the prediction of treatment response with herbal medicine was 9.75 years, which indicated that the children of the younger age group exhibited better treatment response than the children of the older age group. In GH treatment, a previous study attempted to identify the determinants of treatment response by multiple regression analysis14). The importance of the findings of our study lies in the fact that we were able to identify the determinants of response to herbal medicine treatment for height gain in children by multiple regression analysis by adjusting for confounding factors; to our knowledge, the present study is the first such attempt in herbal medicine treatment. Thus, the findings of our study might help guide herbal medicine treatment and clinical assessment.

Several factors are predictive for adult height, including genetic factors (i.e., MPH), malnutrition, systemic disease, and environmental effects21),27),28). The predictors of GH treatment are the dosage of GH, MPH, delayed bone aging, height and weight at birth, age of initiation of therapy, rate of growth in the first year of birth, BMI, and baseline level of insulin-like growth factor I2),13),29–31). In our study, although the occurrence of secondary sex characteristics was not found to affect treatment response, patient age was an important predictive factor. In previous studies, the effect of age on treatment response with herbal medicine was a matter of controversy7),9). However, since neither study adjusted for confounding factors, the results of the present study might be more reliable in comparison. Additionally, patient age was reported as an important predictor of GH therapy response2),30),31). In Korea, the ages of onset of puberty in boys and girls were reported to be 12.1 ± 1.5 years and 11.3 ± 1.3 years, respectively32). The results of subgroup analysis in the present study revealed a significant increase in height percentile in the younger age group, but not in the older age group. Additionally, herbal medicine treatment for height gain was more effective in children below 9.75 years than in those older than 9.75 years, according to our results of ANCOVA. Therefore, we recommend the initiation of herbal medicine therapy at as young an age as possible during the pre-pubescent period, in order to increase the treatment response. However, this result should not be generalized as implying that herbal medicine is ineffective for height gain in older children without considering alternative explanations for the discrepancy in results between the younger and older age groups, such as small sample size, administration of various herbal medicine, and subgrouping of responders.

Treatment for rhinitis was also an important factor in predicting treatment response. At our clinic, we prescribe herbal medicine containing Forsythiae fructus or Schizonepeta tenuifolia for the treatment of rhinitis. A previous study reported that 22% of the children who presented at a KM hospital were affected by rhinitis15). The prevalence of rhinitis among students in general elementary schools during the last 12 months in an urban area of Korea was reported to be about 39% (95% CI, 36.2–41.8)33). It is known that rhinitis interferes with quality of sleep34),35). Longer duration of rhinitis and symptoms of severe rhinitis are correlated with lower sleep quality and lower height percentile36),37). The proportion of children of short stature among a group of children with allergic diseases was reported as being high38). Therefore, Lee et al. suggested that the impairment of GH release caused by sleep disorders in children with rhinitis leads to decreased rates of height gain37). The effect of Forsythiae fructus on growth has not yet been investigated. Forsythiae fructus is usually prescribed for the treatment of rhinitis. Therefore, it may be speculated that the effect of treatment of rhinitis by Forsythiae fructus resulted in enhanced quality of sleep and affected the treatment response for height gain in the present study. Further studies are required to prove this hypothesis.

The results of multiple regression analysis revealed no associations between several independent variables and treatment response. However, because of the lack of statistical power, the negative results of the present study should be interpreted with caution. Heredity are important predictors of growth and GH treatment response13),21). However, MPH was not found to be a predictor of response to herbal medicine treatment in the present study; this is in agreement with the results of previous studies on herbal medicine, which reported no influence of heredity on treatment response6),7). The baseline height percentile was also not found to be a predictor of treatment response in the present study. However, whether baseline percentile is a determinant of treatment response with herbal medicine is a matter of controversy6–9). In a previous study, GH therapy was reported as being effective in children of the lower percentile group30). Therefore, further investigation regarding the influence of baseline height percentile on treatment response with herbal medicine is required. The administration of Ephedra sinica, which is frequently prescribed for the treatment of obesity39),40), was not found to influence treatment response. Additionally, because of the small proportion of children with a history of previous therapy, we were unable to evaluate the influence of previous KM or GH therapy on treatment response in the present study.

Treatment duration was not found to affect the treatment response. Previous studies have reported contradictory results regarding the same6),9). In clinical practice, physicians tend to prescribe herbal medicine for longer durations for patients exhibiting poor treatment response. In cases where the treatment response is determined to be good enough, the treatment duration might be relatively shorter. Therefore, the results of the present study should not be interpreted as indicating the lack of influence of treatment duration on the rate of height gain. This hypothesis may be further verified in a retrospective study by comparison of the influence of treatment duration on the rate of height gain between similar groups using an appropriate matching method. And mean treatment duration of our research was relatively short (105 days) in height growth treatment. Therefore, when interpreting the relationship between treatment duration and treatment response, careful interpretation is necessary using the results of a longer period of treatment.

In the present study, several treatment methods were adopted for height growth, with the dryness-moistening formula (潤燥劑) being the most frequently administered. It is composed of several moistening medicinal herbs used for treatments of conditions of dryness and fluid deficiency22). In TEAM, ‘Yin’ is associated with body fluid or structure. ‘Yin deficiency’ causes a decrease in moisture, which leads to dryness of the body. The dryness-moistening formula restores moisture lost because of ‘Yin deficiency’, causing body fluid or structures to be replenished, thus enhancing height gain. The tonifying and replenishing formula (補益劑) is used to nourish blood or replenish qi22). Both these treatment methods target the deficiency syndrome, which is an important factor in growth treatment in children15). The phlegm-dispelling formula (祛痰劑) is used for dissipating or expelling phlegm. Phlegm is defined as a pathological product accumulated in the body, which causes several diseases22). Several medicinal herbs are used to remove phlegm accumulated in the body. In the present study, Ephedra sinica was administered for weight loss, especially in obese children. Although the influence of herbal formulas for obesity treatment on height growth in children is not well established, it is an interesting subject for further research. The digestant formula (消食劑) is used to enhance digestion and disperse stagnant food22). Herbal medicine treatment for digestive problems is known to aid height growth in children8). Sleep disorders due to chronic rhinitis interferes with growth in children37). The tranquillizing formula (安神劑) is used to relieve mental stress and stabilize the mind to improve sleep quality22). The exterior-releasing formula (解表劑) is used to discharge pathogenic factors from superficial parts of the body such as the nose or skin.22). In our study, several exterior-releasing formulas were used to treat rhinitis.

There are several limitations of our retrospective review. First, the weight and height at birth and the prevalence of atopic diseases were not investigated as potential predictors in this study. In order to address this issue, such variables should be investigated in future studies. Second, in the present study, because of the lack of data on the follow-up BA, the difference in height percentile between baseline and post-treatment was evaluated as the dependent variable. Although the difference in height percentile is widely used in studies on height gain using herbal medicine, the rate of height gain (cm/year)14),41) and expected adult height, calculated based on the BA42), according to the Bayley and Pinneau method43), are also commonly evaluated treatment outcomes. Future studies should evaluate the BA again at the end of treatment in order to determine the expected adult height and also perform annual measurement of height in order to calculate the rate of height gain. Third, because of the small sample size, we were unable to calculate the optimal cutoff ages for the male and female patients separately. We recommend subgroup analysis with lager sample size in future studies in order to address this issue. Fourth, the negative results of the present study should be interpreted with caution considering the possibility of type II errors because of the small sample size. The predictive power of the results of our study is relatively low since other variables that might explain the variability of response between the age groups might have been missed. The final goal of our study was to construct a prediction model for herbal medicine treatment that could help determine the optimal treatment regimen and goals based on realistic expectations14). Prospective long-term follow-up studies with large sample sizes are required to conclusively confirm the findings of our study. Finally, the criteria for inclusion in our study were relatively broad. We did not target specified herbal prescription or medicinal herbs. We could not identify effect of specific prescription or medicinal herbs. However, this study was the first report regarding the determinants of response to various herbal medicine treatment for height gain in children. We recommend further analysis involving a broad spectrum of patients in order to achieve a greater generalization of results. We also recommend study design about effect of specific herbal prescription or medicinal herbs on height growth.

The present study has several strengths. Ours is the first study involving the use of multiple regression analysis for the prediction of treatment response with various herbal medicine for height gain. Consequently, the results of our study are more robust compared to those of previous studies that performed univariate analysis6),7),9),15),37),44). Determination of the cutoff value of treatment age can be helpful in the assessment of the best course of treatment in clinical practice. The relatively wide ranges of inclusion criteria and herbal medicine prescriptions allow the generalization of results of the present study. A prospective cohort study is required to conclusively confirm the feasibility of the determinants of treatment response. However, in consideration of the time and money that a prospective study would entail, retrospective chart reviews should be given precedence. The results of our study will form the basis for further studies in this area.