Abstract
There has been a significant increase in suicide rates in the United States (U.S.) over the past two decades. Studies have highlighted the need for further exploration of suicide risk factors, particularly combinations of factors. In this study, a pharmacovigilance analysis was conducted to better understand drugs and indications as risk factors for suicide using data from the U.S. Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) and Adverse Event Open Learning through Universal Standardization (AEOLUS), a standardized version of FAERS. Association rule mining techniques were applied to 85,071 cases involving suicide-related adverse reactions and demographic subsets of these cases. Preliminary results reveal combinations of drugs and indications that may increase the likelihood of suicide, with certain combinations potentially affecting some demographic groups more than others. Further work is needed to validate the initial findings, explore subpopulations, and determine the broader implications for suicide prevention.
Introduction
Suicide is the 10th leading cause of death nationally.1 With suicide rates increasing nearly 30% from 1999 to 2016, there is a growing need for a comprehensive approach to suicide prevention.2 A key component of this approach is identifying risk factors to better understand and prevent suicidal thoughts and behaviors.3 One realm in which there is still much to be learned is which combinations of drugs or indications lead to an adverse drug reaction (ADR) of suicide and how these combinations may differ across various demographic groups.
With incidence rates of serious ADRs reaching 6.5%, ADRs have been identified as an important area of study.4 The FDA Adverse Event Reporting System (FAERS), a database containing post-marketing reactions for drugs in the United States, has facilitated the study of ADRs.5,6 There is much that can be discovered by mining this database for correlations between specific drugs and suicide-related outcomes. Previous studies using FAERS to investigate drugs and suicide have looked at the relationship between isotretinoin usage and depression and suicide, finasteride usage and suicidal behaviors, and suicidal ideation and suicidal behavior as ADRs for antidepressant drugs.7-9
One challenge in using FAERS is the existence of duplicate cases and misspellings, which makes it necessary to perform extensive data cleaning.10 In 2015, Banda et al. addressed this issue by creating a publicly available version of FAERS that removes duplicate reports, maps events to standardized concept identifiers (drug names to RxNorm concepts and outcomes to SNOMED-CT concepts), and precomputes summary statistics.11 This database, called the Adverse Event Open Learning through Universal Standardization (AEOLUS), was created using FAERS data from January 2004 to June 2015. A limitation of AEOLUS is that it does not preserve the demographic data present in FAERS, so it is necessary to use FAERS as a supplementary resource to retrieve patients’ demographic information.
Through basic analysis of adverse event databases, individual drugs that lead to a specific reaction can be discovered. To discover combinations of drugs that lead to a specific reaction, data mining techniques such as association rule mining (ARM), which generates association rules between items in a database, can be applied.12 An efficient method to find association rules in a database is via the Apriori algorithm.13 Applying ARM to adverse event reporting data can help uncover hidden drug-drug interactions that have serious effects on patients. Previous studies show that applying ARM to FAERS or other spontaneous reporting systems is a promising avenue for knowledge discovery.14 Yildirim applied ARM to FAERS to discover rules associating demographic information with adverse events for cases involving ciprofloxacin.15 Harpaz et al. performed ARM on FAERS data from 2008 to find multi-item associations between drugs and ADRs.16 Guo et al. used ARM to investigate the ADRs of two specific drugs, gadoversetamide and rofecoxib.17
The objective of this study is to apply ARM to cases in FAERS that have a suicide-related ADR in order to determine what combinations of drugs and indications are associated with suicide-related ADRs. Identifying these common combinations could provide guidance for further investigations of the effect of such combinations on suicide risk. Additionally, this preliminary study investigates how these association rules differ between male and female patients as well as across patients in different age groups, which can potentially contribute to informing personalized risk assessment.
Methods
Figure 1 depicts an overview of the study approach, which is based on the Knowledge Discovery in Databases process.18 The four major steps involved are: (1) Data selection to identify cases with a suicide-related reaction, (2) Data processing and subsetting to split the dataset into distinct demographic groups to allow for more detailed analyses, (3) Data mining using ARM, and (4) Interpretation and evaluation of results using visualization methods and comparison to relevant scientific literature. Data extraction, analysis, and visualization were conducted using the Julia general purpose programming language, MySQL database management system, and Tableau.
Figure 1.
Overview of approach.
Data selection
The publicly available AEOLUS database, a standardized form of FAERS, was used as the primary data source. AEOLUS removes duplicate cases and applies standardized vocabulary to map FAERS labels to RxNorm Concept Unique Identifiers (for drugs) and SNOMED-CT identifiers (for indications).11 Additionally, AEOLUS precomputes some basic statistics such as the proportional reporting ratio (PRR). PRR is calculated as PRR = (a/(a+b))/(c/(c+d)), where a is the number of cases with the suspected drug and suspected ADR, b is the number of reports with the suspected drug and without the suspected ADR, c is the number of reports without the suspected drug and with the suspected ADR, and d is the number of reports without the suspected drug and without the suspected ADR.19
The version of AEOLUS used for this study was created using FAERS data from January 2004 to June 2015.11 The format of the FAERS data collected from January 2004 to August 27, 2012 differs slightly from the format for data collected from September 2012 onwards in that the earlier data identify cases using isr, while the more recent data identify cases using primaryid. Consequently, the AEOLUS database keeps track of both isr and primaryid.
The AEOLUS concept table was queried for all concepts with names that match “%suicid%”. This query returned the following concepts: “depression suicidal,” “completed suicide,” “suicidal behavior,” “suicidal ideation,” “suicide attempt,” and “suicide of relative.” The first five concepts were identified as suicide-related ADRs and were chosen to be used to extract relevant cases for this study. “Suicide of relative” was left out because it is not an adverse reaction experienced by patients themselves.
The identifiers (primaryid for cases prior to August 2012; isr for cases after August 2012) of cases with reactions of “depression suicidal,” “completed suicide,” “suicidal behavior,” “suicidal ideation,” and “suicide attempt” were extracted. These identifiers were used to obtain: (1) the list of drugs and (2) the list of indications for patients who experienced one or more of these reactions. The AEOLUS tables used to extract these data were standard_case_drug and standard_case_indication, respectively. The resulting dataset is referred to as the “full dataset.” Basic statistics regarding the drugs involved in these cases were extracted from the AEOLUS table standard_drug_outcome_statistics, which contains precomputed PRRs and frequencies for each drug-reaction combination.
Data processing and subsetting
The full dataset was segmented according to two demographic factors: (1) sex and (2) age. The purpose of data subsetting was to amplify signal detection of factors that increase suicide risk for specific population groups.20,21 AEOLUS does not contain demographic data, but by joining the AEOLUS and FAERS databases using primaryid and isr, demographic data were retrieved for the cases used in this study. Separate lists of drugs and indications were generated for female and male patients who had one of the previously mentioned suicide-related ADRs. Cases without a gender code were excluded. Similar lists were generated for patients in the following age groups: <15, 15-24, 25-34, 35-44, 45-54, 55-64, and >64. The selection of age groups follows the practices of the Centers for Disease Control and Prevention, with the exception that all age ranges less than 15 years old have been combined into one age group because children < 15 years old all tend to exhibit low rates of suicide.22,23 Cases with no age field and cases with ages measured in units other than years were excluded.
Data mining
A Julia package (ARules.jl) was used to perform ARM via the Apriori algorithm for the ten datasets (full dataset and nine subsets).24 The support threshold was set at 0.005 and the confidence threshold was set at 0.01. Support was calculated using the equation supp(X → Y) = P(X ∪ Y), confidence using the equation conf(X → Y) = P(X | Y) = supp(X ∪ Y)/supp(X), and lift using the equation lift(X → Y) = supp(X∪Y)/(supp(X)supp(Y)).25 The maximum rule length was set to six to allow for the discovery of more complex rules should they exist.
Interpretation and evaluation
Tableau Desktop 2018.2 was used to visualize and organize the top drugs, top indications, and association rules discovered in this study. Association rules were grouped by Rhs (right-hand side) to make it easier to see differences between the association rules for each demographic group.
PubMed and Side Effect Resource (SIDER), a centralized database of known ADRs that appear in drug labels, were used as resources to validate findings involving ADRs for individual drugs.26 SIDER 4.1, which was released in October 2015, was the version used in this study. PubMed was also used to validate findings involving indications or combinations of drugs.
Results
The characteristics of patients who had one or more suicide-related ADRs are described in Table 1. In total, AEOLUS contains 85,071 cases that involve a suicide-related ADR, with the most common suicide-related ADRs being completed suicide, suicidal ideation, and suicide attempt. Note that some cases are associated with more than one suicide-related ADR, so the percentages for each ADR add up to more than 100%.
Table 1.
Description of patient cohort used in analysis of suicide-related ADRs.
| # cases | % cases | # cases | % cases | ||
|---|---|---|---|---|---|
| ADR | Age group | ||||
| Completed suicide | 32,337 | 38.01% | < 15 | 7,396 | 8.69% |
| Depression suicidal | 1,062 | 1.25% | 15 - 24 | 8,425 | 9.90% |
| Suicidal behavior | 1,494 | 1.76% | 25 - 34 | 9,774 | 11.49% |
| Suicidal ideation | 32,227 | 37.88% | 35 - 44 | 12,783 | 15.03% |
| Suicide attempt | 21,298 | 25.03% | 45 - 54 | 14,153 | 16.63% |
| Gender | 55 - 64 | 8,901 | 10.46% | ||
| Male | 31,156 | 36.62% | > 65 | 5,768 | 6.78% |
| Female | 39,484 | 46.41% | Non-year unit | 635 | 0.75% |
| Unknown gender | 17,778 | 20.90% | Unknown age | 17,236 | 20.26% |
Table 2 shows the top drugs associated with cases involving suicide-related ADRs. The left-hand side of the table shows the top ten drugs ranked by PRR using a cutoff of count ≥ 3 to highlight stronger signals, while the right-hand side of the table shows the top ten drugs ranked by case count. The “S” superscript following a drug name denotes that the drug has a suicide-related side effect (suicide, suicidal ideation, suicide attempt, completed suicide, or suicidal behavior) in the SIDER database. The “P” superscript denotes that the drug does not have a documented suicide-related side effect in the SIDER database, but there exists at least one PubMed article linking the drug to a suicide-related reaction.
Table 2.
Top drugs associated with cases with suicide-related ADRs ranked by PRR (left) and count (right).
| Top 10 Ranked by PRR | Top 10 Ranked by Count | ||||||
|---|---|---|---|---|---|---|---|
| Drug Name | Reaction | PRR | Count | Drug Name | Reaction | PRR | Count |
| Carbon Black | Completed suicide | 244.91289 | 12 | VareniclineS | Suicidal ideation | 12.08389 | 4069 |
| Carbaryl | Completed suicide | 147.95526 | 3 | Acetaminophen | Completed suicide | 3.58407 | 4025 |
| Atropine / Hyoscyamine / Phenobarbital / Scopolamine Oral Solution | Completed suicide | 134.50738 | 5 | AlprazolamS | Completed suicide | 5.54899 | 3504 |
| Acetaminophen / Dextromethorphan P, 27 / Doxylamine Oral Tablet | Completed suicide | 113.334 | 9 | ParoxetineS | Suicidal ideation | 6.09995 | 2780 |
| Captan | Completed suicide | 112.7289 | 4 | QuetiapineS | Completed suicide | 5.25461 | 2621 |
| Ethylene Glycol | Completed suicide | 112.43552 | 144 | EthanolP,28 | Completed suicide | 58.77839 | 2482 |
| Acetaminophen / Caffeine / Pyrilamine Oral Capsule | Completed suicide | 110.96644 | 3 | Acetaminophen / Hydrocodone Oral Tablet | Completed suicide | 24.60013 | 2410 |
| Helium | Completed suicide | 105.68436 | 5 | VareniclineS | Suicide attempt | 9.58679 | 2216 |
| Konjac Mannan | Completed suicide | 103.57267 | 7 | ZolpidemS | Completed suicide | 3.83457 | 2169 |
| Carbon Monoxide | Completed suicide | 102.87414 | 112 | ClonazepamS | Completed suicide | 3.94694 | 2144 |
S found in SIDER; P, # found in PubMed (where # is the reference to the PubMed article)
Of the 85,071 cases with suicide-related ADRs, 44,782 cases (52.64%) have one or more known indications. 15,617 cases (18.36%) were tagged with only “product used for unknown indication” or “drug use for unknown indication,” while the remaining cases had no indication information. Figure 2 depicts the indications that appear most frequently in the 44,782 cases with known indications. The ten most frequent indications are depicted in their own circles, with the remaining indications grouped together in the circle marked “other.”
Figure 2.
Top indications associated with cases with suicide-related ADRs.
Of the 85,071 cases with suicide-related ADRs, 84,330 (99.13%) contained drug data. Table 3 displays the association rules for these cases. Rules should be interpreted as Lhs (left-hand side) → Rhs (right-hand side). For example, the first row of Table 3 should be read as {Ibuprofen} → Acetaminophen, meaning that among patients who had suicide-related ADRs, those who were taking ibuprofen were likely to be also taking acetaminophen. The left-hand side of the rule can contain more than one element, but the top drug association rules that were discovered all have left-hand sides that are one-element sets. In Tables 3-9, only rules with conf > 0.18 and lift > 2 are displayed. The results are grouped by Rhs and ranked from greatest to least lift value within each Rhs. Support and confidence tend to exaggerate the importance of rules involving items that appear frequently in the database even if the relationship between items is weak, but lift attempts to correct for frequency and is thus chosen as the primary measure of rule interestingness in this study.29
Table 3.
Top association rules for drugs in all cases with suicide-related ADRs, grouped by Rhs and ordered by descending lift.
 |
Table 9.
Top association rules for drugs in cases involving 45-54 years old who experienced suicide-related ADRs, grouped by Rhs and ordered by descending lift.
 |
The association rules for indications in the 44,782 cases with known indications with conf > 0.18 and lift > 2 are as follows: {depression | smoking cessation therapy} → anxiety (conf = 0.458, lift = 6.083), {sleep disorder} → anxiety (conf = 0.279, lift = 3.707), {insomnia} → anxiety (conf = 0.242, lift = 3.209), {anxiety | smoking cessation therapy} → depression (conf = 0.597, lift = 2.698), {anxiety} → depression (conf = 0.490, lift = 2.212). Unlike in Table 3, some of these rules contain left-hand sides that involve more than one element. For example, the first rule implies that among patients who experienced a suicide-related ADR, those who have indications of both depression and smoking cessation therapy are also likely to have an indication of anxiety.
Tables 4 and 5 show the differences between drug association rules for the 39,095 suicide-related ADR cases that involve female patients and contain drug data vs. the 30,035 cases involving male patients. Tables 6 and 7 show indication association rules for the 21,533 suicide-related ADR cases that involve female patients and contain indications data vs. the 17,721 cases involving male patients.
Table 4.
Top association rules for drugs in cases involving female patients with suicide-related ADRs, grouped by Rhs and ordered by descending lift.
 |
Table 5.
Top association rules for drugs in cases involving male patients with suicide-related ADRs, grouped by Rhs and ordered by descending lift.
 |
Table 6.
Top association rules for indications in cases involving female patients with suicide-related ADRs, grouped by Rhs and ordered by descending lift.
 |
Table 7.
Top association rules for indications in cases involving male patients with suicide-related ADRs, grouped by Rhs and ordered by descending lift.
 |
Association rules for drugs were found to differ across age groups. Table 8 displays association rules for the 8,317 cases that have drug data and involve a patient between 15 and 24 years old. Table 9 displays the association rules for the 14,153 cases that have drug data and involve a patient between 45 and 54 years old. These two age groups are particularly worthy of investigation because suicide is a significant cause of death in both of these age groups. In 2016, suicide was the second leading cause of death among 15-24 year old Americans (436 cases) and the fourth leading cause of death among 45-55 year old Americans (8,437 cases).22
Table 8.
Top association rules for drugs in cases involving patients 15-24 years old who experienced suicide-related ADRs, grouped by Rhs and ordered by descending lift.
 |
The only association rule with conf ≥ 0.18 and lift ≥ 2 that resulted from mining the 4,463 cases that have indications data and involve a patient between 15 and 24 years old is {insomnia} → anxiety (conf = 0.280, lift = 4.287). Table 10 displays association rules for the 7,138 cases that have indications data and involve a patient between 45 and 54 years old. Many significant association rules were discovered in this age group, so in order to highlight the most interesting rules, only rules with conf ≥ 0.25 and lift ≥ 4 are displayed.
Table 10.
Top association rules for indications in cases involving patients 45-54 years old who experienced suicide-related ADRs, grouped by Rhs and ordered by descending lift.
 |
Discussion
In this preliminary study, FAERS cases involving suicide-related ADRs were used to generate basic statistics and association rules. ARM was performed on all cases with suicide-related ADRs, cases with male patients, cases with female patients, and cases split into seven different age groups.
Table 3 contains the association rules discovered by performing ARM on all cases with suicide-related ADRs. These rules are good starting points for future studies intending to investigate combinations of drugs that may lead to suicide. One example is the rule {Valproate} → quetiapine (conf = 0.26, lift = 3.36). Valproate and quetiapine have been found to increase the rate of self-harm and suicide events when taken individually, but there is yet to be a study of how suicide rates are affected by taking these two drugs concurrently.30
Tables 4 and 5 illustrate how data segmentation allows for the discovery of rules that are hidden when analyzing all patients as one group. For example, one association rule that is seen in the male population is {Amlodipine} → lisinopril (conf = 0.20, lift = 6.06). Amlodipine and lisinopril are often used in combination to treat hypertension, but the preliminary findings of this study suggest that other drugs could be considered when treating hypertension in male patients in order to mitigate the risk of suicide-related ADRs.31
Additionally, this study highlights sex differences when it comes to indications in cases involving suicide-related ADRs. Tables 6 and 7 show that “pain” appears frequently among the top association rules for indications for female patients (present in six out of 18 rules), but “pain” does not appear at all in the top association rules for indications for male patients. This is potentially due to greater pain sensitivity in females, but these findings nonetheless suggest that females experiencing pain concurrently with other indications present in the association rules in Table 6 could have an elevated risk of suicide-related ADRs.32
Drug association rules also differ across age groups. Table 8 shows that the top rules for drugs in cases involving 15 to 24-year-old patients are very similar to the top rules mined using all patients (Table 3), but Table 9 reveals association rules present in 45 to 54-year-old patients that were not significant in the general population. Examples of such rules include {zolpidem} → alprazolam (conf = 0.20, lift = 2.03), {temazepam} → diazepam (conf = 0.23, lift = 3.83), and {albuterol} → varenicline (conf = 0.40, lift = 3.63). Table 10 reveals groups of indications that may increase the likelihood of suicide-related ADRs for patients that are 45 to 55 years old. This study highlights that certain drugs and indications may have differing effects on the suicide risk of patients in different age groups.
One limitation of this study is that it is difficult to determine cause and effect relationships. In the FAERS and AEOLUS databases, it is difficult to determine whether a drug that is associated with a suicide-related ADR is a drug leading to suicidal behaviors or a drug purposefully taken to perform suicide. For example, acetaminophen appears in several association rules discovered in this study. However, since acetaminophen, more commonly known by its brand name Tylenol, is one of the most commonly abused drugs, it is possible that taking acetaminophen is a manifestation of suicidal behaviors rather than the cause.33 This suggests that some of the drugs that appear in the association rules mined in this study are drugs used in suicide attempts rather than drugs leading to suicide-related behavior. Future studies investigating suicide could supplement data from spontaneous reporting systems like FAERS with longitudinal data from electronic health records and claims data, which would provide better evidence for cause and effect relationships.
Another challenge of this study was determining appropriate minimum support and confidence levels for (1) running the Apriori algorithm and (2) determining which rules outputted by the Apriori algorithm are worthy of further investigation. In this study, metrics such as PRR, support, confidence, and lift were used for ranking individual drugs and associations. Additional metrics such as chi-square could be calculated and compared to assess performance. Since support and confidence tend to overlook rare rules and are thus imperfect measures of rule interestingness, implementing the Apriori algorithm using a different interestingness metric as a threshold is worthy of future exploration.34 Sindhu and Kannan have used the Apriori algorithm modified to utilize PRR instead of the traditional confidence as a threshold to mine FAERS.35 Ibrahim et al. have used a hybrid Apriori algorithm that uses PRR and chi-square as interestingness thresholds to mine FAERS.36 Applying these algorithms to data from different demographic segments, as was done in this study, would be a good next step.
Other next steps include studying how drugs and indications differ among various suicide-related ADRs and conducting formal evaluations. Early findings suggest that certain drugs are more likely to lead to some suicide-related ADRs than other suicide-related ADRs. For example, varenicline is associated with 4,069 cases of suicidal ideation in AEOLUS but only 681 cases of completed suicide, whereas acetaminophen is associated with only 876 cases of suicidal ideation but 4,025 cases of completed suicide. Formal evaluations will involve validating the results with clinical experts and established medical knowledge resources as well as determining if there is a statistically significant difference among demographic groups. For example, quantitative evaluations could involve comparisons of rules to characterize similarities as well as differences (e.g., using F-score or Matthews correlation coefficient). Additionally, in order to determine which combinations of drugs and indications contribute uniquely to suicide-related ADRs, a control group could be used (e.g., by removing association rules that appear in populations without suicide-related ADRs from those generated in this study). Potential applications of these findings for supporting clinical decisions and precision medicine will also be further explored (e.g., using longitudinal electronic health data to examine suicide risk for patients who are prescribed a given combination of drugs).37
Conclusion
There is a need to better understand risk factors for suicide. Using data from a spontaneous reporting system and data mining techniques, this study characterized combinations of drugs and indications associated with suicide-related ADRs. These initial findings can serve as a resource for informing investigations of potential drug-drug interactions or comorbidities for suicide risk. Further work is needed to validate the initial findings, explore subpopulations, and determine the broader implications for suicide prevention.
Acknowledgments
This work was funded in part by National Institutes of Health grant U54GM115677. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Figures & Table
References
- 1.WISQARS leading causes of death reports - CDC. Available at: https://webappa.cdc.gov/sasweb/ncipc/leadcause.html.
- 2.Stone DM, Simon TR, Fowler KA, Kegler SR, Yuan K, Holland KM, et al. Vital signs: Trends in state suicide rates - United States, 1999-2016 and circumstances contributing to suicide - 27 states, 2015. MMWR Morb Mortal Wkly Rep. 2018;67(22):617–24. doi: 10.15585/mmwr.mm6722a1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.McLean J, Maxwell M, Platt S, Harris F, Jepson R. Risk and protective factors for suicide and suicidal behavior: a literature review. Edinburgh: Scottish Government Social Research. 2008 [Google Scholar]
- 4.Lazarou J, Pomeranz BH, Corey PN. Incidence of adverse drug reactions in hospitalized patients: a meta-analysis of prospective studies. JAMA. 1998;279(15):1200–5. doi: 10.1001/jama.279.15.1200. [DOI] [PubMed] [Google Scholar]
- 5.Rodriguez EM, Staffa JA, Graham DJ. The role of databases in drug postmarketing surveillance. Pharmacoepidemiol Drug Saf. 2001;10(5):407–10. doi: 10.1002/pds.615. [DOI] [PubMed] [Google Scholar]
- 6.FDA Adverse Event Reporting System (FAERS) Public Dashboard. Available at: https://www.fda.gov/Drugs/GuidanceComplianceRegulatoryInformation/Surveillance/AdverseDrugEffects/default.htm.
- 7.Gupta M, Vujcic B, Gupta A. Completed suicide in acne patients using isotretinoin: Results from the US FDA Adverse Events Reporting System (FAERS). Journal of the American Academy of Dermatology. 2018 Sep 1;79(3) [Google Scholar]
- 8.Gupta M, Vujcic B, Gupta A. Suicidal behaviors (suicidal ideation, suicide attempt and completed suicide) in patients treated with finasteride for hair loss: Results from the US FDA Adverse Event Reporting System (FAERS) database. Journal of the American Academy of Dermatology. 2018 Sep 1;79(3) [Google Scholar]
- 9.Stübner S, Grohmann R, Greil W, Zhang X, Müller-Oerlinghausen B, Bleich S, Rüther E, Möller HJ, Engel R, Falkai P, Toto S, Kasper S, Neyazi A. Suicidal ideation and suicidal behavior as rare adverse events of antidepressant medication: Current report from the AMSP Multicenter Drug Safety Surveillance Project. Int J Neuropsychopharmacol. 2018 Sep 1;21(9):814–821. doi: 10.1093/ijnp/pyy048. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Sakaeda T, Tamon A, Kadoyama K, Okuno Y. Data mining of the public version of the FDA Adverse Event Reporting System. Int J Med Sci. 2013 Apr 25;10(7):796–803. doi: 10.7150/ijms.6048. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Banda JM, Evans L, Vanguri RS, Tatonetti NP, Ryan PB, Shah NH. A curated and standardized adverse drug event resource to accelerate drug safety research. Sci Data. 2016 May 10;3 doi: 10.1038/sdata.2016.26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Agrawal R, Imieliński T, Swami A. Mining association rules between sets of items in large databases. ACM SIGMOD Record. 1993 Jun 1;22(2):207–216. [Google Scholar]
- 13.Agrawal R, Srikant R. Fast algorithms for mining association rules in large databases. Proceedings of the 20th International Conference on Very Large Data Bases. 1994:487–99. [Google Scholar]
- 14.Wang C, Guo XJ, Xu JF, Wu C, Sun YL, Ye XF, Qian W, Ma XQ, Du WM, He J. Exploration of the association rules mining technique for the signal detection of adverse drug events in spontaneous reporting systems. PLoS One. 2012;7(7) doi: 10.1371/journal.pone.0040561. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Yildirim P. Association patterns in open data to explore ciprofloxacin adverse events. Appl Clin Inform. 2015 Dec 16;6(4):728–47. doi: 10.4338/ACI-2015-06-RA-0076. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Harpaz R, Chase HS, Friedman C. Mining multi-item drug adverse effect associations in spontaneous reporting systems. BMC Bioinformatics. 2010 Oct 28;11 doi: 10.1186/1471-2105-11-S9-S7. Suppl 9:S7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Guo K, Lin H, Xu B, Yang Z, Wang J, Sun Y, Xu K. Detecting potential adverse drug reactions using association rules and embedding models. International Symposium on Bioinformatics Research and Applications. 2017 May 29;:373–378. [Google Scholar]
- 18.Fayyad U, Piatetsky-Shapiro G, Smyth P. The KDD process for extracting useful knowledge from volumes of data. Communication of the ACM. 1996;39:27–34. [Google Scholar]
- 19.Evans SJ, Waller PC, Davis S. Use of proportional reporting ratios (PRRs) for signal generation from spontaneous adverse drug reaction reports. Pharmacoepidemiol Drug Saf. 2001 Oct-Nov;10(6):483–6. doi: 10.1002/pds.677. [DOI] [PubMed] [Google Scholar]
- 20.Umetsu R, Abe J, Ueda N, Kato Y, Matsui T, Nakayama Y, Kinosada Y, Nakamura M. Association between selective serotonin reuptake inhibitor therapy and suicidality: Analysis of U.S. Food and Drug Administration Adverse Event Reporting System data. Biol Pharm Bull. 2015;38(11):1689–99. doi: 10.1248/bpb.b15-00243. [DOI] [PubMed] [Google Scholar]
- 21.Almenoff JS, Pattishall EN, Gibbs TG, DuMouchel W, Evans SJ, Yuen N. Novel statistical tools for monitoring the safety of marketed drugs. Clin Pharmacol Ther. 2007 Aug;82:2157–66. doi: 10.1038/sj.clpt.6100258. [DOI] [PubMed] [Google Scholar]
- 22.Leading Causes of Death by Age Group. Available at: https://www.cdc.gov/injury/wisqars/pdf/leading_causes_of_death_by_age_group_2016-508.pdf.
- 23.Suicide Statistics – AFSP. Available at: https://afsp.org/about-suicide/suicide-statistics/
- 24.Julia Association Rules Package. Available at: https://github.com/bcbi/ARules.jl.
- 25.Hornik K, Grün B, Hahsler M. arules-A computational environment for mining association rules and frequent item sets. Journal of Statistical Software. 2005;14(15):1–25. [Google Scholar]
- 26.Kuhn M, Letunic I, Jensen LJ, Bork P. The SIDER database of drugs and side effects. Nucleic Acids Res. 2016 Jan 4;44(D1):D1075–9. doi: 10.1093/nar/gkv1075. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Modi D, Bhalavat R, Patterson JC. 2nd. Suicidal and homicidal behaviors related to dextromethorphan abuse in a middle-aged woman. J Addict Med. 2013 Mar-Apr;7(2):143–4. doi: 10.1097/ADM.0b013e318281a547. [DOI] [PubMed] [Google Scholar]
- 28.Dinis-Oliveira RJ, Carvalho F, Duarte JA, Dias R, Magalhães T, Santos A. Suicide by hanging under the influence of ketamine and ethanol. Forensic Sci Int. 2010 Oct 10;202(1-3):e23–7. doi: 10.1016/j.forsciint.2010.04.047. [DOI] [PubMed] [Google Scholar]
- 29.Wright A, Chen ES, Maloney FL. An automated technique for identifying associations between medications, laboratory results and problems. J Biomed Inform. 2010 Dec;43(6):891–901. doi: 10.1016/j.jbi.2010.09.009. [DOI] [PubMed] [Google Scholar]
- 30.Hayes JF, Pitman A, Marston L, Walters K, Geddes JR, King M, Osborn DP. Self-harm, unintentional injury, and suicide in bipolar disorder during maintenance mood stabilizer treatment: A UK population-based electronic health records study. JAMA Psychiatry. 2016 Jun 1;73(6):630–7. doi: 10.1001/jamapsychiatry.2016.0432. [DOI] [PubMed] [Google Scholar]
- 31.Cappuccio FP, Markandu ND, Singer DR, MacGregor GA. Amlodipine and lisinopril in combination for the treatment of essential hypertension: efficacy and predictors of response. J Hypertens. 1993 Aug;11(8):839–47. doi: 10.1097/00004872-199308000-00011. PubMed PMID: 8228208. [DOI] [PubMed] [Google Scholar]
- 32.Fillingim RB, King CD, Ribeiro-Dasilva MC, Rahim-Williams B, Riley JL 3rd. Sex, gender, and pain: a review of recent clinical and experimental findings. J Pain. 2009 May;10(5):447–85. doi: 10.1016/j.jpain.2008.12.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.El Majzoub I, El Khuri C, Hajjar K, Bou Chebl R, Talih F, Makki M, Mailhac A, Abou Dagher G. Characteristics of patients presenting post-suicide attempt to an Academic Medical Center Emergency Department in Lebanon. Ann Gen Psychiatry. 2018 May;25:17–21. doi: 10.1186/s12991-018-0191-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Lallich S, Teytaud O, Prudhomme E. Association rule interestingness: Measure and statistical validation. Quality measures in data mining. 2007:251–275. [Google Scholar]
- 35.Sindhu MS, Kannan B. Detecting signals of drug-drug interactions using association rule mining methodology. IJCSIT) Int J Comput Sci Inf Technol. 2013;4(4):590–4. [Google Scholar]
- 36.Ibrahim H, Saad A, Abdo A, Sharaf Eldin A. Mining association patterns of drug-interactions using post marketing FDA’s spontaneous reporting data. J Biomed Inform. 2016 Apr;60:294–308. doi: 10.1016/j.jbi.2016.02.009. [DOI] [PubMed] [Google Scholar]
- 37.Prince JD. Precision Medicine: An Introduction. Journal of Electronic Resources in Medical Libraries. 2017 Oct 2;14(3-4):120–9. [Google Scholar]