Stack Sequential Algorithm to Analyse Adverse Drug Reaction

Stack Sequential Algorithm to Analyse Adverse Drug Reaction Amiya Kumar Tripathy, Nilakshi Joshi, Aslesha More, Divyadev Pillai, Amruni Waingankar Abstract— An adverse drug reaction (ADR) is ‘a response to a medicine which is noxious and unintended, and which occurs at doses normally used in human beings’. During the last decades it has been estimated that such adverse drug reactions (ADRs) are the 4th to 6th largest cause for mortality in different countries. They result in the death of several thousands of patients each year, and many more suffer from ADRs. The percentage of hospital admissions due to adverse drug reactions in some countries is about or more than 10%. In addition suitable services to treat ADRs impose a high financial burden on health care due to the hospital care of patients with drug related problems. Some countries spend up to 15-20% of their hospital budget dealing with drug complications. The existing scenario is manual, costly, not easily portable and the findings are not reported to the responsible authorities in timely manner. To overcome these flaws of the existing system, we propose a automated ADR detecting  system. This is an interactive system platform for detecting ADR from the specified combination of drugs. If ADR is detected , the system will suggest some appropriate combination of drugs which will solve the specified purpose. The detection part of ADR is done by using algorithms like Chi-Square, PRR (Proportionality Result Ratio), Combinational.(Which are already implemented). The solution part i.e, suggesting appropriate drug combination is implemented by Sequential decoding algorithms and the stack sequential algorithm. Keywords—Adverse Drug Reaction, Adverse Drug Event, Bayesian Confidence Propagation Neural Network, Information Component, Frequent Pattern, Online Analytical Processing Database , Operating System, Random Access Memory, Medicines and Healthcare pro ducts Regulatory Agency, World Health Organization. I. INTRODUCTION An adverse drug reaction (ADR) is an injury caused by taking medication.[1] ADRs may occur due to a single dose or long term administration of a drug or result from the combination of two or more drugs ,as this last expression might also imply that the effects can be beneficial.†ADE can be mainly caused from medication errors. A serious adverse event is any event that: Is fatal Is life-threatening Is permanently/significantly disabling Requires or longterm hospitalization Causes birth defects Requires intervention to prevent permanent impairment or damage II. RELATED WORK Study of design elements include exploration of relevant multiple outcomes (utilization and/or safety), sample size calculations, cohort accrual procedures, and the timing and method of data collection. The custom questionnaires can include those related risks potentially possible and identified, also missing information in RMPs or can be designed to address specific regulatory issues. Single data capture or multiple data capture phases enable abstraction of clinical information from medical chart review by prescribers responsible for treatment initiation in primary care, over a time frame relevant to study needs analysis plans can be tailored to address novel analytical issues and also convey thoughtful, appropriate, and comprehensive analysis of the data. Study reports have been prepared with scientific rigor to provide brief or in depth presentation of results relevant to the products safety and efficiency [4]. Before a medicine is granted a license it has to go through the strict tests and routine checks to ensure that it is acceptably safe and effective. All the medicines which are effective, can cause adverse drug reactions which in general term we say a side effect, which can range from a very minor occurrence to being very serious. For a medicine to be authorized licensed, the benefits of the medicine must satisfy all the possible conditions of the medicine causing adverse effects in patients. Many a times, it is not easy to identify if the side effect is due to a medicine, or something else. Even if it is only a suspicion that a medicine or combination of medicines has caused a side effect, the patients or doctors are asked to send the proper reports of the symptoms and drugs prescribed, to FDA. Reports received on suspected side effects are evaluated, with the information like clinical trial data, medical literature or data from international medicines regulators, for identifying previously unidentified safety issues or side effects[5]. In statistics, a confidence interval (CI) is a population parameter estimation and it indicates the reliability of an estimate. It is calculated from the observations, in principle differs from sample to sample. The frequently observed interval contains the parameter and is determined by the confidence level or confidence coefficient. Confidence intervals contains a range of values (interval) that act as best estimates of the unknown population parameter. However, in infrequent cases, none of these values may cover the value of the parameter. The level of confidence of the confidence interval would indicate the probability that the confidence range captures this true population parameter given a distribution of samples. The confidence interval contains the parameter values that, when tested, should not be rejected with the same sample. Greater levels of variance yield larger confidence intervals, and hence less precise estimates of the parameter. Confidence intervals of difference parameters not containing 0 imply that there is a statistically significant difference between the populations. III.SYSTEM ARCHITECTURE The database comprises of a combination of drugs-symptoms and the adverse drug reactions associated with them in particular. The database is a heterogeneous aggregation of the demographic, combination, drug, symptoms and reactions of ADRs. It previously consists of Information related to the drug description and attributes of the patient as well. After evaluating the patient info which involves the symptoms and the medicines the doctor has prescribed to that patient the working system would provide a safe case to the end user. These cases are consistent and can be further used by the Doctor/Pharmacist to prescribe the combinations. These safe cases would be ago ahead for the doctor about his prescription to that particular patient. These are tested cases and determine the validity of that doctor’s prescription. The probability of occurrence of an ADR and its detection for a new patient is our main goal. The input after being assessed derives a result of an ADR case. The occurrence of ADR means that the Doctor has to modify his combination of Drugs accordingly and convert an ADR case to a safe case. This should result in the doctor making changes that prove to be safe to the patient according to our system. The input given by the Doctor detecting an ADR case is then matched with the database in the available system. The Database already comprises of ADR cases found out, so when a current patient whose symptoms and the prescribed medicine by the doctor matches it leads to a ADR detection. Figure 1: System Architecture The doctor will give input to the system including the patients symptoms and the medicines prescribed by the doctor to that current patient .These information are the important factors for the detection of the ADR through the system. IV.SYSTEM FLOW Figure 2: System flow The System flow involves extracting the Database which includes the ADR cases and the safe cases. These cases are stored in the database and comprises of the drug-symptom combination. When a Doctor provides input to the system with the Symptoms and drugs, the system evaluates the validity using the PRR algorithm and provides a value which is then matched with the current records in the database of ADRs. In occurrences of safe case the doctor continues with his combination of medicines prescribed, but in the event of an ADR that is occurred the doctor realizes that a medical negligence could occur. To prevent such an occurrence he modifies his drug combination for the safety of the patient. V. EXPERIMENTAL SETUP For making experimental setup we have developed the structured database in XAMPP with MySQL and we have made front end in HTML for executing various queries. The database which we are using is the official database released by food and Drug Administration (FDA)[2]. The database is updated after every three months. Database consist complete information about Demographic data of patients, drug, indications, reactions, therapy ETC. data of 5000 patients was considered as test cases. The database which we have contains following: Drug information; Reaction information; Patient outcome information; Information on the source; Figure 3: database Screenshot This table shown below gives complete statistics about the test cases of adverse drug reaction received in various years. The test cases are reported by various hospitals, doctors, pharmacist, clinical researches, and drug manufacturers. This table shows the number of reports received by Food and Drug Association and entered into FDA Adverse Event Reporting System by type of report since the year 2003 until the end of the second quarter of 2012. Table 1: database statistics Table 1 provides information regarding the data that is present in the database[2]. Each year shows a number of new cases that were registered with the US FDA. Expedited Cases involve those which were reported as soon as it was detected. Direct Cases involve those which were reported by individuals/ independent medical practitioners. Non-expedited cases are those cases which were reported much after occurrence. The number of cases that were received by US FDA were much higher than those that were entered. Details regarding registered cases is available until the second quarter of the year 2012 only. VI. METHODS AND ALGORITHMS Proportionality Reporting Ratio (PRR) The PRR algorithm is a statistical method which is used to detect ADRs in Electronic health records and databases .The working of this algorithm relies on the fact that when an ADR (related to a particular event) is identified for a medicinal product (say medicinal product P), this adverse event is relatively reported more often in association with this product P than with any other products in the database. This gradual increase in the reporting of events for the medicinal product P in consideration is reflected in the table below based on the total number of cases stored in the database. Table 2: Contingency matrix for PRR In the table mentioned the elements calculated are the individual available cases in the available database .Therefore a given individual case may contribute to only a single cell of the table, where the cases refer to the multiple products or the adverse events[7] PRR= A/(A+B) C/(C+D) The general criteria to run the PRR are as follows: Value A is the number of cases with the defective medicinal product P involving an adverse event R. Value B is the number of cases related to the defective medicinal product P, involving any other adverse events but R. Value C is the number of cases involving event R in relation to any other medicinal products but P. Value D is the number of cases involving any other adverse events but R and any other medicinal products but P The system performs the calculations of the PRR on all the case counts instead of the ADRs to be chosen to keep the individuality between the variables used to calculate PRR so that the difference of the PRR will not be underestimated. The calculation of the PRR is done as follows: For evaluating given cases of nausea involving medicinal product ‘allopurinol l’ = 15% (e.g. 15 reports of diarrhea amongst a total of 100 reports reported with medicinal product ‘allopurino l’). For evaluating a given number of reports of nausea with other medicinal products in a database = 5%. Thus, the Proportionality Reporting Ratio is equal to 3 . The chi-square (χ2) statistics The Chi-square is a statistic, which is traditionally used in dis proportionality analyses. The Chi-square is used as an alternative measure of association between the medicinal product P and the adverse event R based on the following calculation:[8] X2= (AD-BC)2(A+B+C+D)/[(A+B)(C+D)(A+C)(B+D)] When the PRR is displayed with the X2 statistic : The PRR ≠¥ 2 The X2 ≠¥ 4 The number of indivisual cases greater or equal to 3. Stack Sequential We introduce the drug combination optimization algorithms and show how they relate to the algorithms used in sequential decoding. Fully factorial datasets, where every possible drug combination is tested, grow exponentially with the number of drugs (n). See Text S1 for the relevant equation and an example dataset. In computational terms we say that the complexity is O(an). The O-notation indicates the order of growth of an algorithm basic operation count as a function of the input size. An exponential growth is not practical for large n, therefore our aim is to find algorithms with improved efficiency, for example with a linear dependency on n, expressed as O(n). The problem of finding the optimal estimate of the encoded sequence is described as a walk through a tree. To appreciate the analogy with the search for the optimal drug combination, the tree shown in Figure 4 can be compared with the trees used in one of the original descriptions of the stack sequential algorithm [14]. An alternative version of the tree, the ‘‘trellis’’ depiction shown in Figure 5, eliminates nodes representing redundant drug-dose combinations. The stack is a sorted list of all examined combinations (best on top). Notations : A1= BACITRACIN A2= KADCYLA B1= ERYTHROMYCIN B2= RAMIPRIL S1 – the process initially contains only the list of the measurements in the absence of any drug (the root of the tree of Figure 4). Figure.4 Tree representation of the data S2 The parsing begins from the top of the sorted list. After the search completes it moves one level up in the braches of figure 4. Combinations already used are ignored for future extensions. S3 – Once the combination reaches its maximum size , the parsing ends. This is similar to reaching the top of the tree of Figure 4. Since we consider the best combination, instead of best path, we do not delete any combination from the processed list. When we find a combination been already used, we move to the next combination in the sorted list. We do not combine different doses of the same drug with each other, to limit the size of the search, but this is not an essential feature as shown in Figure 2,. The algorithm is efficient in searching combinations in which the outcome is not purely additive, because it overcomes non-linearities by backtracking to nodes in the tree. Figure.5 Trellis-like representation showing combination of the data. S1 Examine all drugs based on strength, doses and rank. S2 The best drug combination is saved from the processed list. S3 Select the best single drug and call it Cbest. S4 Take the Combination of Cbest with all other drugs, increasing the drug size by 1, measure the biological scores, and store the list of drugs of this size. At this step the algorithm moves one level upwards in the tree of Figure 4. S5 If the new combinations scores better than Cbest, this combination is used as the new Cbest and return to previous step. If no new combination scores better than Cbest, backtrack to the next best combination in the previous size, mark it as Cbest and return to the previous step. S6 Backtrack value should be limited to a specific value. S7 Repeat S4 to S6 till we find that the maximum size for the combinations is reached. VII. CONCLUSION In this system, by using PRR in association with Chi-Square, an attempt has been made to help and assist the doctors/pharmacists to perform safe drug evaluation. An experimental study using test cases and combinations from a doctor was performed and the results obtained were very promising. The system proposes a unique method for correcting the prescribed combination of drugs in case of an ADR event occurrence using Stack Sequential. The possibility of vast Patient Record data available allows for extracting the results available to the system. The approach used in this paper can be to provide an impetus and improve existing systems that provide detect Adverse Drug Reactions. In the field of Pharmaceutical and medical diagnosis, there is always the scope for uncertainty. This system has been built to provide a naà ¯ve and safe understanding of the drug combinations on the experience of doctors only, so there will always be a scope for ambiguous or uncertain diagnosis. The developed system does not give a 100% accurate results as not even the doctors can claim to do so; however, its results are promising. It can be used as a tool to complement the doctors’ knowledge and could assist them to reach a conclusion. The system will give the doctor an upper hand to decide whether to use the results evaluated from the algorithm and prevent an ADR. By using this system, many essential results can be obtained, thus reducing the effects of wrong prescriptions to some extent. With the support of various medicinal and pharmaceutical practitioners and hospitals, higher probability of getting the positive results right can be obtained. With an extensive database of medical records to mine from, this could be useful to build helpful medical assistance software that can be of great use to all doctors and pharmacists using this system. The system will also help the medical fraternity in the future by helping them in providing safer medical assistance to the patients and doctors. VIII. REFERENCES [1] Search Algorithms as a Framework for the Optimization of Drug Combinations Diego Calzolari1., Stefania Bruschi1., Laurence Coquin1, Jennifer Schofield1, Jacob D. Feala2, John C. Reed1, Andrew D. McCulloch2, Giovanni Paternostro1,2* 1 Burnham Institute for Medical Research, La Jolla, California, United States of America, 2 Department of Bioengineering, University of California San Diego, La Jolla, California, United States of America. [2] Safety of Medicines A guide to detecting and reporting adverse drug reactions. [3] Statistical methods for knowledge discovery in adverse drug reaction surveillance G. Niklas Norà ©n. [4] Adverse drug reactions: definitions, diagnosis, and management I Ralph Edwards, Jeffrey K Aronson [5] MANUAL OF OPERATIONS ADVERSE DRUG REACTIONS ONLINE REPORTING SYSTEM (ADORS) [6] An Information Technology Architecture for Drug Effectiveness Reporting and Post-Marketing Surveillance -Surendra Sarnikar , Amar Gupta, Ray Woosley(2006). [7] A multi-agent intelligent system for detecting unknown adverse drug reactions through communication and collaboration -Ayman Mohammad Mansour Wayne State University(2012) [8] ADRTrace: Detecting Expected and Unexpected Adverse Drug Reactions from User Reviews on Social Media Sites Andrew Yates, Nazli Goharian [9] Detect adverse drug reactions for the drug Pravastatin. Yihui Liul ‘Institute of Intelligent Information Processing,’ Shandong Polytechnic University, China(2012)