You are on page 1of 4

Informatics in Medicine Unlocked 15 (2019) 100169

Contents lists available at ScienceDirect

Informatics in Medicine Unlocked


journal homepage: www.elsevier.com/locate/imu

Treatment of peptic ulcers with a Siddha medicine, “Sirucinni Uppu” and T


prediction with regression models
Kumaravel Appavooa,∗, Mudiganti Ram Krishna Raob
a
School of Computing, India
b
Bharath Institute of Higher Education and Research, Department of Industrial Biotechnology, 173, Agaram Road, Tambaram, 600073, Chennai, India

A R T I C LE I N FO A B S T R A C T

Keywords: Application of computational tools in medical science can assist physicians with analysis of disease. Herein,
Siddha prediction based on a subset featured approach with the Siddha medical treatment dataset is utilized for peptic
Peptic ulcer ulcers, using a simple linear regression (LR) based model as predictor, and measuring its effectiveness via error
Acalypha fruticosa estimates and statistical significances. The herbal salt of Acalypha fruticosa (Siruccini Uppu) is dissolved in dis-
Sirucinni uppu
tilled water and administered to peptic ulcer patients in standard doses with honey before diet. The results were
Gummam
Prediction
tabulated to show efficacy of treatment. Clinically ‘Sirucinni uppu’ is shown as an effective medicine in treating
Linear regression ‘Gunmam’ (peptic ulcer) patients with 80% satisfactory result, 10% fair, and 10% moderate result. The present
Gaussian processes analysis is beneficial in the effective usage of patterns and relationships reflected in datasets collected from real
Error estimate cases of peptic ulcer disease treatment. This approach of using such algorithms for early diagnosis of peptic
ulcers could be employed by physicians to treat these patients more effectively.

1. Introduction treatment. This paper presents a prediction model based on a medical


treatment dataset, an approach by using a simple linear regression (LR)-
Peptic ulcers are a prevalent disease, and each year peptic ulcer based model as predictor, and measures its effectiveness through error
disease (PUD) affects four million people around the world [1]. The estimates and statistical significance. The advantages of applying LR
pathophysiology of PUD involves an imbalance between offensive fac- over other models are its simplicity and interpretability for establishing
tors (Acid, Pepsin, and H. pylori) and defensive factors (Mucin, Pros- the correlation between each feature and the prognosis [3,4]. The us-
taglandins, Bicarbonate, Nitric oxide and Growth factors). Considering ability of LR as a base learner is evaluated by performing experimental
the several side effects (impotence, gynaecomastia and haematopoeitic analysis with a dataset consisting of a sample cohort of small size (60
changes) of modern pharmcologic agents, indigenous drugs possessing instances), and its effectiveness is established based on significant
fewer side effects might better treat peptic ulcers sans side effects. The performance. The aim of the present study is therefore to develop a
clinical studies for the efficacy of herbal salt extracted from ‘Acalypha prediction model which can assist physicians in decision-making while
fruticosa’ (HSAF) on gastric ulcers was found. HSAF was effective on reducing false negative cases.
human subjects with peptic ulcer without perceptible side effects.
During the last few decades, various computational tools have been 2. Model background
shown to be of value to medical practitioners in evaluating, diagnosing
and treating patients in a more effective and realistic way. This com- The objective of this study is to generate a regression model with
bination of medical science and computational science has provided statistically significant methods for the prediction of the prognosis of
excellent value for predicting the disease [2]. The early and accurate peptic ulcer with standard dosage practice in Siddha, a form of alter-
diagnosis and prediction of peptic ulcer can effectively reduce mor- native medicine. The uniqueness of such alternative forms of medicinal
bidity and mortality due to chronic conditions such as perforation, practice include their cost effectiveness, easy of availability, and time-
bleeding, and onset of malignancy. Thus, an accurate and reliable tested authenticity. The database for peptic ulcer cure was procured
technique is necessary, with predictive capabilities. Our findings in this from the work of Rao et al, 2014 [5]. The data of sixty patients were
regard are beneficial in the development of effective usage of patterns taken for the present study. The time in days for the relief or cure of
and relationships, reflected in datasets collected with real cases of PUD symptoms of peptic ulcer patients, such as epigastric pain, flatulence,


Corresponding author.
E-mail addresses: drkumaravel@gmail.com (K. Appavoo), mrkrao1455@gmail.com (M.R. Krishna Rao).

https://doi.org/10.1016/j.imu.2019.100169

Available online 26 March 2019


2352-9148/ © 2019 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/BY-NC-ND/4.0/).
K. Appavoo and M.R. Krishna Rao Informatics in Medicine Unlocked 15 (2019) 100169

indigestion, loss of appetite, vomiting, sore tongue and constipation Criteria (exclusion): Complications of peptic ulcer including he-
were recorded. The patients were from both sexes and of different age morrhage, perforation, gastric outlet obstruction, radiating abdominal
groups. The dose of Sirucinni uppu was administered ranging from pain as can occur in pancreatitis, appendicitis, acute abdominal colic,
130 mg to 500 mg B.D. with honey before diet, and the study period cancer of the stomach, gall stones, hiatal hernia, cirrhosis of the liver,
varied from 7 days to 15 days, and to a maximum duration of 40 days. and jaundice excluded patients from the study, because such patients
The clinical experimental protocol for ‘Shirucinni uppu’ (AJ/IAEC/10/ would require invasive treatment.
12) was approved by the CPCSEA/IAEC of Mohamed Sathak A.J College Criteria for withdrawal: Irregular medication, irregular follow-
of Pharmacy, Sholinganallur, Chennai, India. ups.
Routine examination and assessment: The full details of patient
3. Proposed algorithm history and physical examination of the patients was recorded as per
the Proforma. The clinical assessment was done initially at the end of 4
The main method applied in the present report, namely linear re- days, 7 days, 14 days and 21 days follow-up.
gression, addresses the problem of estimating the output value based on Incidence of PUD is higher in male (64%) than female The occur-
inputs. Here the main output is prognosis, along with other outputs rence was 26% in age group 40 to 49; 18% between 50 and 69% and
including flatulence and epigastric pain as symptoms for peptic ulcer. 14% between 20 and 39 years. Duration of treatment varied from 8
The linear model to be fit must try to minimize the sum of squares error days to 28 days depending upon patient condition. The average treat-
(deviations between the data points and the regression line) is re- ment period was 12–18 days. Improvement from the disease is recorded
presented by from the history of prognosis, which showed relief of different symp-
toms, as follows:
y = c0 + c1x1 + … ….+ cnxn

By determining the regression coefficients c0, c1, …cn, the re- i. 52% of patients relieved of gastric pain and nausea in three days,
lationship between the output attribute, y, and the input variables x1, 32% in 4 days, 12% in five days, and 4% in six days. 56% of patients
… …, xn can be estimated [6–9]. obtained relief from flatulence in four days, 30% in three days, 8%
in five days and 6% in six days.
ii. 20% of patients were relieved of indigestion in three days, 42% in
4. Dataset description
four days, 22% in five days, 14% in six days and 2% in seven days.
In total, 80% of patients reported satisfactory improvement, and
The symptoms playing the roles of attributes in the data set are
20% improved fairly. Ten PUD patients were admitted and provided
explained briefly here.
treatment. Of these, 50% were male and 50% female, aged between
Epigastric pain: Peptic ulcer can cause pain in the upper part of the
19 and 68 years. The patients improved in a period six to eleven
stomach just below the breast bone and often just after taking food.
days from their symptoms, and were then discharged. 60% of pa-
Flatulence: The feeling of bloating of the stomach leading to dis-
tients were relieved of epigastric pain and indigestion in four days,
comfort and difficulty in breathing. This can also cause belching.
30% in five days and 10% in three days. 60% of patients were re-
Indigestion and Loss of appetite: Usually peptic ulcers are formed by
lieved of flatulence in five days, 20% in four and 20% in three days.”
excessive release of stomach acid or due to the presence of a bacterium,
Helicobactor pylori. The ulcer can cause discomfort, leading to improper
Fig. 1 indicates the system architecture for the proposed study, in
digestion and loss of appetite.
Vomiting: Due the excessive acidity in the stomach, vomiting is a
common phenomenon in peptic ulcer patients.
Sore tongue: As the digestion is disturbed in peptic ulcer patients,
their tongue develops inflamed taste buds, manifested as small painful
bumps and red color. This condition is highly irritable, and the patient
feels discomfort in chewing, swallowing and speaking.
Constipation: Constipation is one of the major effects of peptic ulcer,
due to a disturbed and unregulated digestion process. The bowel
movement of patients becomes irregular, which can lead to further
complications to the digestive system and the general physiology.
The dataset instances were generated from the records filled during
the treatment procedure and duration as follows. The data and de-
scription of peptic ulcer treatment with SU were followed from Rao
et al, 2014, [5].

5. Study participants

Both men and women, and members of all the races and ethnic
groups, were eligible for this trial. Treatment was administered on an
Inpatient/Outpatient basis. The sample size was 60 patients, and these
were selected from the Arignar Anna Government Hospital of Indian
Medicine and Homeopathy, Chennai-600006.

6. Criteria for selection

Criteria (inclusion): patients with epigastric pain, heart burn, re-


gurgitation, nausea/vomiting, loss of appetite, abdominal discomfort,
flatulence, indigestion, sore tongue and constipation, which were signs
of initial stages of peptic ulcer, were included for this treatment. Fig. 1. System architecture for the proposed study.

2
K. Appavoo and M.R. Krishna Rao Informatics in Medicine Unlocked 15 (2019) 100169

Table 1
Statistical description of the Eight attributes in the dataset for peptic ulcer
treatment.
Sl. No Symptom Maximum Minimum Mean Standard
Deviation

1 Prognosis 28 5 13.1831 5.003


2 Epigastric Pain 6 3 3.75 0.836
3 Flatulence 6 3 3.967 0.802
4 Indigestion 7 3 4.37 0.988
5 Improved Appetite 7 3 4.283 0.976
6 Relief from 6 3 3.733 0.8
Vomiting
7 Relief from Sore 6 3 4.267 0.778
tongue
8 Relief from 7 3 4.633 0.78
Constipation
Fig. 2. Linear relationship between patient instance and prognosis.

Table 2
Linear regression for prognosis and its related symptoms.
Symptoms Regression Line Coefficient of
Determination

Epigastric Pain Y = 0.004 x +3.611 0.008


Flatulence Y = 0.013 x +3.542 0.091
Indigestion Y = 0.002 x +4.289 0.001
Improved appetite Y = 0.001 x +4.319 0
Stoppage of Vomiting Y = 0.003 x +3.618 0.006
Normalized Sore Tongue Y = 0.006 x +4.059 0.023
Recovery from Constipation Y = 0.003 x +4.535 0.005

Table 3
The Error estimates for peptic ulcer symptoms.
Relief of Linear Regression Gaussian Processes
Fig. 3. Linear relationship between patient instance and flatulence. Symptoms
Mean Absolute Root Mean Absolute Root
Error Squared Error Squared

Error Error

Epigastric Pain 0.7074 0.8403 0.7917 0.9754


Flatulence 0.6119 0.8217 0.5946 0.8059
Indigestion 0.8222 1.009 0.8418 1.0254
Appetite 0.7889 0.9877 0.7802 0.9992
Vomiting 0.6467 0.9723 0.674 0.8161
Sore Tongue 0.6921 0.8061 0.6988 0.802
Constipation 0.651 0.7997 0.6659 0.8001

Table 4
p-values for significant symptoms.
Symptom Regression Coefficient p-value∗

Flatulence 0.301662 0.02065


Prognosis 0.288097 0.026531

Fig. 4. Comparison of error estimates (mean absolute errors) by linear regres- (*p < 0.05).
sion and Gaussian process.
clustering, development of association rules, and visualization. Fig. 2
which the stages of experiment are shown. This flowchart shows the indicates the scatter diagram of the features in the patient dataset for
method based on mainly two components in the sequence - firstly the prognosis. Fig. 3 indicates features in the patient dataset for flatulence.
linear regression aspect, followed by error handling contributed by a In both Figs. 2 and 3, the relationship is evident, irrespective of order of
Gaussian process. These are applied on the dataset, and treatment re- patient instances shown on the X-axis. Both figures depict the range in
cords were collected from the hospital as mentioned before. The figure number of days for elimination of symptoms, and show the data points
shows the iteration progress until the error is within threshold. Here with peaks and troughs for recovering from symptoms such as flatu-
since our data size is small, not only is the error small, but the method lence, as well as for the overall prognosis. Fig. 4 compares error esti-
terminates after completion of all records for classification. The results mates (mean absolute error, root mean square error) by linear regres-
of this method are shown in Tables 2–3, using the open access software sion and Gaussian process. to show the sensitivity of the obtained
tool Weka [10]. Weka is a software package written in JAVA code, and model.
is a common data mining tool for preprocessing, classification, Table 1 shows the statistical description of the dataset for peptic
ulcer treatment, which is expressed as summaries of statistical

3
K. Appavoo and M.R. Krishna Rao Informatics in Medicine Unlocked 15 (2019) 100169

descriptions including the maximum, minimum, mean, and standard the article.
deviation for the cohort of 60 patients. The range of number of days in
which the cure of symptoms takes place provides salient information. Conflicts of interest
Table 2 depicts the outcome of regression analysis to predict prog-
nosis, and another seven symptoms based on the treatment of peptic This is to declare that no conflicts of interest exist among the au-
ulcer for the selected cohort. This table demonstrates the nature of thors.
fitting of the regression line for the dataset with observations X = {xi },
each xi being the symptom and y is the prognosis. Table 3 indicates the Funding
error estimates for peptic ulcer symptoms. This is further to the Linear
Regression (LR) model, with usual standard error estimates MAE and The work is self-stipendiary and no external funding was availed.
RSE are tabulated applying Gaussian processes. In this table we find
that the (GP) performance was better than simple LR. Moreover, it Acknowledgements
shows minimum values for the attribute, i. e. flatulence. Furthermore,
all of the above tables show the absence of an outlier problem for this The authors wish to acknowledge the assistance rendered by all in
dataset. Table 4 shows the p-values of significant symptoms, i. e. the successful competition of the present work.
flatulence and prognosis.
References
7. Discussion
[1] Zelickson MS, Bronder CM, Johnson BL, Camunas JA, Smith DE, Rawlinson D, et al.
Predicting the response to individual drugs based on the symptoms Helicobacter pylori is not the predominant etiology for peptic ulcers requiring op-
eration. Am Surg 2011;77:1054–60.
has proved to be a statistically easier approach, even though the models [2] Bello G, Dumancas GG, Gennings C. Development and validation of a clinical risk-
obtained may contain noise specific to the selected dataset, leading to assessment tool predictive of all-cause mortality. Bioinf Biol Insights 2015;9:1–10.
less accurate results with independent datasets. Therefore, more gen- [3] Opper M, Winther O. Gaussian processes for classification: mean-field algorithms.
Neural Comput 2000;12(11):2655–84.
eralized analytical methods for improved prediction with larger data- [4] Williams CKI. Prediction with Gaussian processes: from linear regression to linear
sets is proposed for a future project. To address the error minimization, prediction and beyond. In: Jordan MI, editor. Learning in graphical models. Kluwer
a Gaussian Processes would be useful, based upon probability theory, to Academic; 1998. p. 599–621.
[5] Rao Ram Krishna Mudiganti, Ganesan A, Sundari G Renga, Kumar M Sathish. The
determine the nonlinear relationship between the attributes implied by curative role of herbal salt of Acalypha fruticosa Forssk.(Sirucinni uppu) salt on peptic
the covariance, since knowledge of one reduces uncertainty in the other ulcer patients. Der Pharm Lett 2014;6(4):44–51.
[11,12]. A similar work was reported by Koc et al., 2017 on bleeding [6] Steyerberg EW. Clinical prediction models. A practical approach to development,
validation, and updating. New York: Springer; 2009.
peptic ulcers [13]. Rezaei et al., 2014 have employed a Fuzzy inference
[7] Mukaka MM. A guide to appropriate use of Correlation coefficient in medical re-
system to assess the severity of peptic ulcer presentation [14]. The re- search. Malawi Med J 2012;24(3):69–71.
gression model described herein shows that [1,2], i. e. Flatulence and [8] Choi Edward, Bahadori Mohammad Taha, Schuetz Andy, Stewart Walter F, Sun
Prognosis, respectively, are statistically significant according to the p- Jimeng. Doctor ai: predicting clinical events via recurrent neural networks. Machine
learning in health care. MLHC; 2016. 2016.
value (Table 4), which is less than 0.05. For other symptoms, it in- [9] Dumancas GG, Adrianto Indra, Bello Ghalib, Dozmorov Mikhail. Current develop-
dicates that there is insufficient evidence in the sample dataset to ments in machine learning techniques in biological data mining. Bioinf Biol Insights
conclude that non-zero correlation actually exists. This predictive 2017;11. 1177932216687545.
[10] https://www.cs.waikato.ac.nz/ml/weka/.
model will assist physicians in the following way: [11] Lasko Thomas A. Efficient inference of Gaussian process modulated renewal pro-
cesses with application to medical event data. Proceedings of the thirtieth con-
a. The physician can expect the relief of a particular symptom in a ference on uncertainty in artificial intelligence (UAI). July 2014.
[12] Rasmussen CE, Williams CKL. Gaussian Processes for machine learning. MIT Press;
patterned duration by treatment with Sirucinni uppu. 2006. ISBN 026218253X@2006MIT.
b. If this prediction fails for more than one symptom, the physician can [13] Koc M, Yoldas O, Kilic YA, Gocmen E, Ertan T, Dizen H, Tez M. Comparison and
understand that the importance of more invasive diagnostic validation of scoring systems in a cohort of patients treated for perforated peptic
ulcer. Langenbeck's Arch Surg 2007;392:581–5.
methods without losing time. This can help in serving the patient
[14] Rezaei Kianaz, Hosseini Rahil, Mazinani Mahdi. A fuzzy inference system for as-
better to obtain an early remedy. sessment of the severity of the peptic ulcers. Computer science & information
technology (CS & it). In: Nagamalai Dhinaharan, editor. ACITY, WiMoN, CSIA,
AIAA, DPPR, NECO, InWeS – 2014; 263–271, 2014. © CS & IT-CSCP2014. https://
Ethical statement
doi.org/10.5121/csit.2014.4527.

The authors have abided by all ethical parameters while preparing

You might also like