Also available as PDF (figures and tables appear in the PDF).
Adolescent suicide attempts represent a critical public health concern shaped by interacting behavioral, interpersonal, and institutional risk factors. This study examined multilevel determinants of suicide attempts among 30,226 secondary school adolescents in the Northern Samar Division of the Philippines using a quantitative cross-sectional design aligned with the World Health Organization (WHO) Global School-based Student Health Survey (GSHS) framework [1]. Data were collected through an anonymous structured online survey assessing substance use, dietary behaviors, peer and family connectedness, school engagement, lifestyle regulation, and bullying victimization. Hierarchical binomial logistic regression was conducted to estimate adjusted associations, with model performance evaluated using deviance statistics, pseudo-R², and area under the receiver operating characteristic (ROC) curve. Suicide attempts were reported by 14.10% of respondents. The final model demonstrated strong explanatory capacity (Nagelkerke R² = .259) and good discrimination (AUC = .795), with overall classification accuracy of 75.3%. Substance-related variables showed significant associations, particularly lifetime cannabis use (OR = 5.56, p < .001), frequent alcohol use (OR = 1.85, p < .001), and high sugar-sweetened beverage consumption (OR = 1.81, p < .001). Relational factors were among the strongest predictors, with adolescents reporting rare or no parental understanding showing nearly threefold increased odds (OR ≈ 3.0, p < .001). Lifestyle dysregulation variables, including short sleep duration (<5 hours; OR = 1.93, p < .001) and high sedentary time (OR = 1.68, p < .001), were also significant. Bullying victimization, particularly cyberbullying (OR = 2.19, p < .001), independently predicted suicide attempts. These findings support an integrated vulnerability model in which substance exposure, relational disconnection, behavioral dysregulation, and victimization jointly elevate suicide risk. Multicomponent school-based prevention strategies addressing these domains are essential for reducing adolescent suicide attempts.
Keywords: adolescent suicide attempts; school-based mental health; bullying victimization; substance use; parental connectedness; hierarchical logistic regression; adolescentsuicideattempts; school-basedmentalhealth; bullyingvictimization; substanceuse; parentalconnectedness; hierarchicallogisticregression
parentalconnectedness;hierarchicallogisticregression. 1.Introduction regulation, impulsivity)interactwithproximalstres- sors(e.g.,bullying,conflict,academicstrain)topre- Suicideremainsamajorglobalpublichealthconcern, cipitatesuicidalbehavior.Empiricalevidenceincreas- withmorethan720,000deathsannually,anditranks inglysupportsthecentralroleofrelationalconnected- amongtheleadingcausesofdeathinadolescentsand nesswithintheseframeworks.Schoolconnectedness youngadults[2]. Foreverysuicidedeath, thereare hasbeenshowntomoderatetheassociationbetween manymoreattempts,andapriorattemptisoneofthe mentalhealthsymptomsandsuicideattemptsinearly strongestpredictorsoffuturesuicidemortality[2].Ado- adolescents,withstrongerconnectednessattenuating lescentsinlow-andmiddle-incomecountries(LMICs) riskeveninthepresenceofdepressiveoraggressive bearadisproportionateburden,accountingfornearly symptoms[7].AnalysisofArgentina’s2018GSHSdata three-quarters of global suicide deaths [2]. Within demonstratedthatschool,parental,andpeerconnect- thiscontext,population-basedsurveillanceplatforms ednesssignificantlymitigatedtheassociationbetween suchastheGlobalSchool-basedStudentHealthSur- bullyingvictimizationandsuicidalbehaviors[8].These vey(GSHS)provideessentialepidemiologicalinfras- findingsalignwiththebelongingnesscomponentof tructureforidentifyingmodifiablebehavioralandpsy- theInterpersonalTheoryofSuicideandsuggestthat chosocialdeterminantsofsuicideattemptsinschool- connectednessoperatesnotmerelyasacorrelatebut attendingyouth[1].TheGSHSwasdesignedtoassess as a buffering mechanism. Large-scale surveillance clustered health-risk behaviors among adolescents datafromhigh-incomesettingsreinforcethatsuicide acrossmultipledomains—includingsubstanceuse,vi- attemptsamongadolescentsareshapedbyinteracting olenceandinjury,sedentarybehavior,physicalactiv- interpersonal,behavioral,andinstitutionalfactors.In ity,socialsupport,andmentalhealth—usingstandard- theUnitedStates,the2023YouthRiskBehaviorSur- izedschool-basedsampling[1].AnalysesofGSHSdata vey reported that 9.5% of high school students had fromdiverseLMICsettingsconsistentlydemonstrate attemptedsuicideinthepreceding12months,with thatsuicidalideationandattemptsco-occurwithbul- higherprevalenceamongfemalestudentsandsexual lying,loneliness,substanceuse,physicalfighting,and minorityyouth[9]. Schoolconnectedness, parental reducedparentalorpeersupport[3,4]. UsingNepal monitoring,andsupportiveenvironmentswereconsis- GSHSdata,Pandeyetal.[3]foundthatloneliness,bul- tentlyassociatedwithlowerprevalenceofsuicideat- lyingvictimization,andalcoholuseweresignificantly temptsacrossdemographicgroups[9].Cross-national associatedwithsuicideattemptsamongadolescents. analyses from the Health Behaviour in School-aged Amulti-countrySoutheastAsianGSHSanalysissimi- Children(HBSC)studyshowthatsuicideattemptsre- larlyreportedthatbeingfemale,bullied,seriouslyin- mainasignificantconcernacrossmultipleEuropean jured,physicallyfighting,lonely,andhighlysedentary countries,withstrongassociationsobservedbetween wereassociatedwithsuicidalbehaviors,whileparental attemptsandbullyingvictimization,loneliness,sub- andpeersupportdemonstratedprotectiveeffects[4]. stanceuse,andlowfamilysupport[10,11].InNordic Thesefindingsreinforcetheneedformultilevelmodel- andWesternEuropeancohorts,schoolconnectedness ingapproachesthatcaptureco-occurringbehavioral and parental communication have been repeatedly andrelationalrisksratherthanexaminingsingleex- identifiedasprotectiveagainstsuicideattemptseven posuresinisolation. Theoretically,suicideattempts afteradjustingfordepressivesymptomsandsocioe- inadolescencearebestunderstoodwithinintegrated conomicfactors[11,12]. InAsia,adolescentsuicide vulnerabilityframeworks. TheInterpersonalTheory attemptsaresimilarlyembeddedwithinrelationaland ofSuicidepositsthatsuicidalbehavioroccurswhen behavioral risk clusters. Multi-country analyses us- thwartedbelongingnessandperceivedburdensome- ingWHO-GSHSdataacrossSouthandSoutheastAsia ness co-exist with acquired capability for self-harm report12-monthsuicideattemptprevalenceranging [5]. Acquired capability develops through repeated between 3% and 13%, with consistent associations exposuretopainfulandprovocativeexperiences,lead- betweenattemptsandbullying,physicalinjury,sub- ingtoincreasedpaintoleranceandfearlessnessabout stanceuse,loneliness,andsedentarybehavior[3,4]. death[5].Bullyingvictimization,physicalinjury,and AregionalanalysisofeightSoutheastAsiancountries substance use may plausibly contribute to this pro- foundthatbeingbullied,seriouslyinjured,physically cessbyincreasingexposuretofear-inducingorpain- fighting,andreportinglowparentalsupportweresig- relatedexperiences.Complementingthismodel,the nificantlyassociatedwithsuicideattempts,whilepeer stress–diathesisframeworkconceptualizessuicidalbe- andfamilyconnectednessdemonstratedprotectiveef- haviorastheproductofstressorsactivatingunderlying fects[4].Thesefindingsareconsistentwithemerging biologicalorpsychologicalvulnerabilities[6]. Inthis datafromEastAsiansettingsshowingthatinterper- framework,distalvulnerabilities(e.g.,emotionaldys- sonalstressorsandacademicpressureinteractwith includedinthefinalfullyadjustedmultivariablemodel behavioral dysregulation to elevate suicide attempt (Model5).
risk among adolescents [13]. In the Philippine con- text,adolescentsuiciderepresentsapersistentlydocu- 2.2 ParticipantsandDataCollection mented public health emergency. National data in- ParticipantswereadolescentsenrolledinGrades7to dicate that 404 students died by suicide during the 12inparticipatingsecondaryschools.Thesurveywas 2021–2022schoolyear,with2,147additionalsuicide administeredelectronicallythroughasecureonline attemptsrecordedwithinthesameperiod[14]. The platform during scheduled school hours. Students PhilippineDepartmentofEducationreportedninesui- completedthequestionnaireusingschool-provided cidedeathsamongstudentsintheNorthernSamarDi- orpersonaldigitaldevicesunderteachersupervision. visionincalendaryear2025.Philippine-specificstud- Participationwasvoluntaryandanonymous.Thesam- iesusingGSHSandcommunity-baseddatasetscon- plecomprisedbothmaleandfemalestudentsacross sistentlyidentifybehavioral,interpersonal,andinsti- earlytolateadolescence(approximatelyages12to20 tutionalriskfactorsascentralcontributorstoadoles- years),withrepresentationfromrural(barangay)and centsuicideattempts[15, 16]. Thecriticalshortage urban(townproper)schoolsacrossGrades7through ofschoolguidancecounselors—withadocumented 12. Bothregularstudentsandthosewithalternative ratio of one counselor per 13,400 students in pub- enrollmenthistories(e.g.,repeater,balik-aral)werein- lic schools, far below the recommended 1:500 ratio cludedinthefinalsampleofN=30,226.
[17]—substantiallyconstrainstheinstitutionalcapac- ityforearlyidentificationandintervention,compound- ingexistingstructuralvulnerabilities.Accordingly,the 2.3 Measures presentstudyappliedamultilevelanalyticframework DependentVariable Suicideattemptswereassessed alignedwithWHO-GSHSdomainstoexaminedeter- usingthestandardizedGSHSmentalhealthmodule minantsofsuicideattemptsamongschool-attending item: "Duringthepast12months, howmanytimes adolescents. GuidedbytheInterpersonalTheoryof didyouactuallyattemptsuicide?"[1].Responseswere Suicide [5] and the stress–diathesis model [6], this dichotomizedforanalysis(0=noattempt;1=atleast study tested whether substance and dietary behav- oneattempt),consistentwiththeanalyticapproach iors,peerandfamilyconnectedness,bullyingvictim- usedinpriorPhilippineGSHSstudies[15,16]. ization,physicalinjury,sedentarybehavior,sleepdu- ration, physical activity, and school engagement in- IndependentVariables Independentvariableswere dependentlyandjointlypredictsuicideattempts.By groupedintofiveconceptualdomainsbasedonthe integratingthesedomainswithinasinglehierarchical finalregressionmodel.Allitemsweredrawnfromstan- logisticregressionframework,thestudyaimedtopro- dardizedGSHSquestionnairemodules[1]: videtheory-informed,surveillance-drivenevidenceto supportcomprehensiveschool-basedsuicidepreven- 1. SubstanceandDietaryBehaviors: Alcoholuse tionstrategiesinthePhilippinesandcomparableLMIC (past30days),lifetimeepisodesofdrunkenness, settings.
sugar-sweetenedbeverageconsumption(past7 days),andlifetimecannabisuse.
2. PeerandFamilyConnectedness: Numberof 2.Methodology close friends and parental or guardian under- 2.1 StudyDesign standingofproblemsandworries(past30days). Thisstudyemployedaquantitative,cross-sectionalan- 3. School Engagement: Missed classes without alyticdesigntoexaminefactorsassociatedwithsuicide permission(past30days)andPhysicalEduca- attemptsamongschool-attendingadolescents.Data tion(PE)attendancefrequency.
werecollectedthroughastructuredonlinesurveyad- 4. Physical Activity and Lifestyle Regulation: ministeredtosecondaryschoolstudentsintheNorth- Days physically active for at least 60 minutes ernSamarDivisionoftheDepartmentofEducation, (past 7 days), daily sedentary time outside of Philippines.Thesurveyinstrumentwasalignedwith schoolworkandsleep,andaveragenightlysleep theGSHScoremodulestoensurestandardizedmea- duration.
surementofhealthbehaviors,interpersonalconnect- 5. BullyingandVictimization: Seriousinjurydue edness, school engagement, lifestyle practices, and tobullying(past12months),bullyingonschool bullyingexperiences[1]. Theanalysisfocusedexclu- property (past 12 months), and cyberbullying sivelyonsuicideattemptswithinthepast12months (past12months).
asthedependentvariableandexaminedpredictors 2.4 StatisticalAnalysis Participants(N=30,226) Datawerecleanedandscreenedpriortoanalysis.De- Characteristic Frequency %ofTotal scriptive statistics were computed to summarize re- SchoolLocation spondent profiles and the distribution of suicide at- Rural(Barangay) 20,928 69.20% temptsandallpredictors. Weightedprevalenceesti- Urban(Townproper) 9,298 30.80% mateswerecalculatedwhereapplicabletoreflectthe Total 30,226 100% complextwo-stageclustersampledesignoftheGSHS GradeLevel andtoproducepopulation-representativeestimates Grade12 2,784 9.20% forschool-attendingadolescents[1]. Binomiallogis- Grade11 2,128 7.00% Grade10 6,720 22.20% ticregressionanalysiswasconductedtoestimatethe Grade9 5,661 18.70% adjustedassociationsbetweenindependentvariables Grade8 6,177 20.40% andsuicideattempts[18].Ahierarchicalmodelingap- Grade7 6,756 22.40% proachwasapplied,culminatinginModel5,thefully Total 25,314 100% adjustedmodelincludingallbehavioral,interpersonal, Sex school-related,lifestyle,andbullyingvariablessimul- Female 17,639 58.40% taneously,consistentwiththemultilevelanalyticstrat- Male 12,587 41.60% egyusedinpriorGSHS-basedstudiesofadolescent Total 30,226 100% suicideattempts[19,20].Estimateswerereportedas AgeGroup log-oddscoefficientsandconvertedtoadjustedodds 20andabove 227 0.80% ratios(AORs)with95%confidenceintervals.Statistical 18–19 1,504 5.00% significancewassetatp<.05.Modelfitwasevaluated 16–17 7,192 23.80% usingdeviancestatistics,AkaikeInformationCriterion 14–15 13,032 43.10% (AIC),BayesianInformationCriterion(BIC),andpseudo- 12–13 8,271 27.40% Total 30,226 100% R2measures[18].
4PsBeneficiaryStatus No 17,185 56.90% 2.5 EthicalConsiderations Yes 13,041 43.10% Thestudyadheredtoethicalstandardsforresearch Total 30,226 100% involving minors and was aligned with GSHS data EnrollmentStatus usepolicies,whichmandatevoluntaryparticipation, Balik-aral 547 1.80% anonymity,andprotectionofrespondentprivacy[1]. Regularstudent 29,269 96.80% Institutional approval was secured before data col- Repeater 410 1.40% lection.Participationwasvoluntary,responseswere Total 30,226 100% anonymous,andnopersonallyidentifyinginformation wascollected. Studentswereinformedofthesensi- successivemodeldemonstratedimprovedfitoverits tivenatureofsuicide-relatedquestionsandwerepro- predecessor. Model1(substanceanddietarybehav- videdinformationonavailableschoolcounselingand iors)yieldedR2McF=.092,R2N=.129,2(14)=2,258.704. supportservices,consistentwithethicalguidelinesfor Model2(peerandfamilyconnectednessadded)im- school-basedsuicideresearchinvolvingadolescents provedtoR2McF=.131,R2N=.182, 2(21)=3,225.496. [21].
Model 3 (school engagement added) yielded R2McF = .140, R2N = .193, 2(32) = 3,444.058. Model 4 (phys- 3.Results ical activity and lifestyle regulation added) yielded Ahierarchicalbinomiallogisticregressionanalysiswas R2McF=.159,R2N=.218,2(40)=3,905.890. Thefully conductedtoexaminefactorsassociatedwithsuicide adjustedModel5(bullyingandvictimizationadded) attemptsamong30,226adolescents.Suicideattempts achievedthestrongestfit: deviance=19,878.90,AIC werereportedby14.10%ofrespondents.
=19,978.90,BIC=20,394.72,R2McF=.192,R2N=.259, 2(49)=4,711.930,p<.001.AICandBICdeclinedmono- 3.1 HierarchicalModelFit tonicallyfromModel1toModel5,confirmingthateach Modelfitwasevaluatedusingdeviance,AIC,BIC,Mc- additionaldomainimprovedpredictiveparsimonybe- Fadden’spseudo-R2(R2McF),Nagelkerke’spseudo-R2 yondthecostofaddedparameters[18].
(R2N),andtheomnibuslikelihoodratiochi-squaretest [18].Allanalyseswereconductedonaweightedsam- pleofN=30,226.AspresentedinTable2,allfivemod- elswerestatisticallysignificant(allp<.001),andeach 3.2 SequentialModelComparisons Sequentialmodelcomparisonswereconductedusing likelihoodratiochi-squaredifferenceteststoevaluate whethereachadditionalpredictordomainproduced statisticallysignificantimprovementoverthepreced- ingmodel[18]. AspresentedinTable3,allfourpair- wisecomparisonswerestatisticallysignificant(allp< .001).Theadditionofpeerandfamilyconnectedness variables(Model2vs.Model1)yieldedthelargestincre- mentalimprovement,2(7)=966.792,p<.001,followed bytheadditionofbullyingandvictimizationvariables (Model5vs.Model4),2(9)=806.039,p<.001. 3.3 OmnibusLikelihoodRatioTests Omnibuslikelihoodratiotestswerecomputedforeach predictorinthefullyadjustedmodeltoassessitsinde- pendentcontributionaftercontrollingforallothervari- ables[18].AsshowninTable4,all14predictorswere
squarevalueintheentiremodel,2(4)=547.900,p<.001. Sleepdurationrankedthirdoverall,2(4)=276.431,p <.001.Amongvictimizationpredictors,cyberbullying producedthestrongestassociation,2(3)=211.264,p< .001.
3.4 ClassificationPerformanceandPredictiveAc- curacy At the optimal cut-off of 0.15—the point of sensitiv- ity–specificityintersection—thefinalmodelachieved overallaccuracyof.753,specificityof.766,andsensitiv- ityof.670(Table5).Thesub-defaultcut-offismethod-
predictedprobabilitiesbelowthatboundary[18]. cantlyhigheroddsofsuicideattempts(OR=1.85,p< TheROCcurve(Figure2)showedclearseparationfrom .001). Lifetimedrunkennessof10ormoreepisodes thediagonalreferenceline,andtheAUCwas.795.By was associated with elevated odds (OR = 1.88, p = establishedbenchmarks,AUCvaluesbetween.70and .001).Cannabisuseexhibitedthestrongestsubstance- .80indicateacceptablediscriminationandvalues .80 relatedassociation;adolescentsreporting10ormore indicateexcellentdiscrimination[18], placingthefi- lifetimeuseshadmorethanfivefoldincreasedodds nalmodelattheupperboundaryofacceptableand (OR=5.56,p<.001).Frequentconsumptionofsugar- approachingexcellentpredictiveperformance. sweetened beverages was also associated with in- creasedoddsinagradedpattern,withthosedrinking 3.5 LogisticRegressionEstimatesbyPredictorDo- threeormoretimesperdayshowing1.81timeshigher main odds(p<.001).
3.5.1 SubstanceandDietaryBehaviors Substance-related variables showed strong associ- 3.5.2 PeerandFamilyConnectedness ations with suicide attempts (Table 6). Compared Relationalfactorsdemonstratedrobusteffects(Table withnon-drinkers,adolescentswhoconsumedalco- 7).Adolescentswithfewerclosefriendshadelevated Model Deviance AIC BIC R2McF R2N 2 df p 1 22,332.12 22,362.12 22,486.87 0.092 0.129 2,258.70 14 <.001 2 21,365.33 21,409.33 21,592.29 0.131 0.182 3,225.50 21 <.001 3 21,146.77 21,212.77 21,487.21 0.140 0.193 3,444.06 32 <.001 4 20,684.93 20,766.93 21,107.91 0.159 0.218 3,905.89 40 <.001 5 19,878.90 19,978.90 20,394.72 0.192 0.259 4,711.93 49 <.001 2 2 2 Note.R2McF=McFadden’spseudo-R2;R2N=Nagelkerke’spseudo-R2.AllmodelsestimatedonN=30,226.
Model(From) Model(To) 2 df p 1 2 966.792 7 <.001 2 3 218.562 11 <.001 3 4 461.832 8 <.001 4 5 806.039 9 <.001 oddsrelativetothosewiththreeormorefriends.Per- 4.1 ModelPerformanceandPredictiveAdequacy ceivedparentalunderstandingshowedastronggradi- Thehierarchicallogisticregressionresultssupportthe enteffect:comparedwithadolescentswhoreportedal- utilityofamultilevelframeworkforexaminingadoles- waysbeingunderstoodbyparentsorguardians,those centsuicideattempts.Sequentialmodelcomparisons whoreportedrarelyorneverbeingunderstoodhadap- confirmedthateachpredictordomaincontributedsta- proximatelythreefoldhigheroddsofsuicideattempts tisticallysignificantincrementalfitbeyondpriorspec- (ORs 3.0,p<.001).
ifications(allp<.001),andthefullyadjustedModel 5achievedthestrongestoverallspecificationacross 3.5.3 SchoolEngagement allindices. ThemonotonicallydecliningAICandBIC Indicatorsofschooldisengagementwereassociated acrossallfivetransitionsconfirmthatsuccessiveim- withincreasedrisk(Table8).Missingschoolforsixor provementsreflectedgenuineexplanatoryincrements more days without permission was linked to higher rather than overfitting [18]. The R2McF of .192 ap- odds (OR = 1.33, p = .002). Lower levels of physical proachesthe.20thresholdindicativeofexcellentfitin activity were associated with increased odds, while logisticregression[22],andthecomplementaryR2N reducedparticipationinPhysicalEducationshowed of.259furtherreflectsmeaningfulexplanatorygain; modestriskelevations.
bothvaluesareconsistentwiththosereportedincom- parablemultilevelGSHS-basedanalysesofadolescent 3.5.4 LifestyleRegulation suicideattempts[19,20,23].
Lifestyleregulationvariablesdisplayedcleargradients (Table9). Greatersedentarytime(5–6or7hoursper Thelargestincrementalimprovementwasproducedby day) was associated with substantially higher odds therelationalconnectednessdomain(2(7)=966.792), (ORs=1.68and1.62,respectively;bothp<.001).Short underscoringthecentralityofparentalandpeersup- sleepdurationwasstronglyassociatedwithsuicideat- portasprotectivefactors—consistentwiththeInterper- tempts;adolescentssleepingfewerthanfivehoursper sonalTheoryofSuicide[5]andGSHS-basedevidence nighthadnearlytwicetheoddscomparedwiththose thatrelationalconnectednessisamongthestrongest sleepingnineormorehours(OR=1.93,p<.001). modifiabledeterminantsofadolescentsuicideattempt risk[19,20,24]. Thevictimizationdomainproduced 3.5.5 BullyingandVictimization thesecond-largestincrement(2(9)=806.039),reflect- Victimizationvariableswereindependentlyassociated ing the well-documented role of bullying and inter- with suicide attempts (Table 10). Being bullied on personal violence in elevating risk through cumula- schoolpropertysixormoretimeswasassociatedwith tivepsychologicaldistress[23,15].Lifestyleregulation 1.70timeshigherodds(p<.001),whilefrequentonline contributedsubstantially(2(8)=461.832),consistent bullying(sixormoretimes)wasassociatedwithmore withevidencethatsleepandsedentarybehaviorinde- thandoubledodds(OR=2.19,p<.001).Experiencing pendentlypredictsuicidaloutcomes[23],whileschool seriousinjuryduetobullyingfourormoretimeswas engagementyieldedthesmallestbutstillsignificant alsosignificantlyassociatedwithincreasedodds(OR= increment(2(11)=218.562)[20,17].
TheAUCof.795providesthemostrobust,threshold- independentindicatorofdiscriminativeperformance. 4.Discussion Byestablishedbenchmarks,thisapproachestheex- Predictor 2 df p Alcoholuse(past30days) 33.925 4 <.001 Lifetimedrunkenness 59.934 3 <.001 Sugar-sweetenedbeverageconsumption(past7days) 76.463 4 <.001 Lifetimecannabisuse 435.829 3 <.001 Numberofclosefriends 107.220 3 <.001 Parentalorguardianunderstandingofproblemsandworries 547.900 4 <.001 Missedclasseswithoutpermission(past30days) 32.745 3 <.001 Daysphysicallyactivefor60minutes(past7days) 73.589 4 <.001 PhysicalEducation(PE)attendancefrequency 17.073 4 0.002 Dailysedentarytime(outsideschoolworkandsleep) 109.350 4 <.001 Averagenightlysleepduration 276.431 4 <.001 Seriousinjuryduetobullying(past12months) 58.114 3 <.001 Bullyingonschoolproperty(past12months) 69.715 3 <.001 Cyberbullying(past12months) 211.264 3 <.001
Accuracy Specificity Sensitivity AUC 0.753 0.766 0.670 0.795 Note.AUC=areaundertheROCcurve.Cut-offvaluesetto0.15(sensitivity–specificityintersection). cellent discrimination threshold of .80 [18], and is behavioral and psychosocial correlates, supporting consistentwiththeclassificationcapacitytypicalof theuseofintegratedmultilevelframeworksinbothre- population-levelbehavioralsurveillancemodels[19, searchandprevention.Substanceusebehaviorswere 20]. Asensitivityof.670atthe0.15thresholdmeans stronglyassociatedwithincreasedsuicideattemptrisk. approximatelyone-thirdoftrueattempterswouldbe Alcoholexposurewaslinkedtohigheroddsofsuicide missed, underscoring that the model is suited for attempts,consistentwithevidencethatalcoholcon- population-levelepidemiologicalinferenceratherthan sumptionimpairsemotionalregulation,increasesim- individual-levelclinicalscreening.
pulsivity,andweakensbehavioralcontrolamongado- lescents[20,16]. Multi-countryanalysesusingGSHS 4.2 Multiple Factors Associated with Suicide At- datasimilarlydemonstratethatalcoholuseissignifi- temptsamongAdolescents cantlyassociatedwithsuicideattempts,particularlyat higherexposurelevels[19,34].Cannabisuseshoweda Thefindingscollectivelyindicatethatmultiplebehav- particularlystrongrelationshipwithsuicideattempts, ioral, relational, school engagement, lifestyle, and consistentwithregionalandPhilippineevidenceiden- victimization factors are associated with suicide at- tifyingcannabisexposureasanimportantbehavioral temptsamongadolescents.Theseresultsextendasub- predictorofsuicidalbehavior[20,16].Thesepatterns stantialbodyofinternationalandregionalevidence canbeinterpretedthroughthestress–diathesisframe- demonstratingthatadolescentsuicidalbehaviorisnot work, which explains that suicidal behavior results a unidimensional phenomenon but arises from the fromtheinteractionbetweenunderlyingvulnerabil- convergenceofinteractingriskandprotectivefactors itiesandenvironmentalstressors,withsubstanceuse acrossbehavioralandpsychosocialdomains[25,26, functioningasbothamaladaptivecopingmechanism 27, 28]. In the Philippines, suicide attempts among and a disinhibiting factor [6]. Family-based longitu- school-attendingadolescentsrepresentapersistently dinal research further demonstrates that substance documentedconcern,withnationaldatashowingsta- abusedisorders,alongsidementalhealthconditions bleandelevatedprevalenceacrosssuccessivesurvey and adverse family environments, constitute some cohorts [29, 30]. Philippine-specific studies consis- ofthemostrobustpredictorsofearly-onsetsuicidal tentlyidentifybehavioral,interpersonal,andinstitu- behavior[35,36]. Theassociationbetweenfrequent tionalriskfactorsascentralcontributorstoadolescent sugar-sweetenedbeverageconsumptionandsuicide suicideattempts[16,31],underscoringtheneedfor attemptsmayreflectbroaderbehavioralriskcluster- evidence-based,school-embeddedpreventionframe- ingamongadolescents,asunhealthydietarybehav- worksadaptedtothelocalcontext[17,32,21].Liuet iorsoftenco-occurwithsubstanceuse,psychological al.[33]furtherdemonstratedthroughasystematicre- distress,andotherhealth-riskbehaviors[1,20]. Be- viewandmeta-analysisthatsuicideandnonsuicidal yondbehavioralexposures,relationalconnectedness self-injuryinchildrenandadolescentssharecommon Predictor Category OR 95%CI p Alcoholuse(past30days) (Reference:0days) 1–2days 1.37 [0.97,1.95] 0.076 3–5days 1.24 [0.83,1.86] 0.293 6–9days 1.45 [0.85,2.49] 0.176 10+days 1.85* [1.31,2.62] <.001 Lifetimedrunkenness (Reference:0times) 1–2times 1.28 [0.87,1.87] 0.208 3–9times 1.03 [0.66,1.51] 0.944 10+times 1.88* [1.29,2.75] 0.001 Sugar-sweetenedbeverages(past7days) (Reference:Didnotdrink) <1time/day 1.11 [0.94,1.30] 0.214 1time/day 1.26* [1.07,1.46] 0.002 2times/day 1.28* [1.09,1.50] 0.002 3+times/day 1.81* [1.55,2.11] <.001 Cannabisuse(lifetime) (Reference:0times) 1–2times 1.35 [0.82,2.20] 0.237 3–9times 1.80 [0.98,3.33] 0.431 10+times 5.56* [3.45,8.93] <.001 Note.*p<.05.OR=adjustedoddsratio;CI=confidenceinterval.
Predictor Category OR 95%CI p Closefriends (Reference:3ormore) 2friends 1.63* [1.33,2.01] <.001 1friend 1.13 [0.90,1.43] 0.386 0friends 1.63* [1.33,2.01] <.001 Parentalorguardianunderstanding (Reference:Always) Mostofthetime 1.00 [0.85,1.19] 0.955 Sometimes 1.52* [1.36,1.71] <.001 Rarely 2.99* [2.62,3.42] <.001 Never 2.97* [2.63,3.36] <.001 Note.*p<.05.OR=adjustedoddsratio;CI=confidenceinterval. emergedasacriticalfactor.Limitedpeerrelationships ity,andhelp-seekingbehavior[20,24].Family-based wereassociatedwithhighersuicideattemptrisk,con- longitudinalresearchfurtherdemonstratesthatthe sistentwithinternationalevidenceshowingthatado- absenceofsupportiveparentalrelationshipsisamong lescentswithfewerclosefriendsexperiencereduced thestrongestinterpersonalpredictorsofearly-onset socialsupportandincreasedsocialisolation[19,20]. suicideattempts[36].Thesefindingsaligndirectlywith Philippine studies using GSHS datasets similarly re- theInterpersonalTheoryofSuicide,whichhighlights portthatweakpeerconnectednessisassociatedwith thatinterpersonaldisconnectionandunmetbelong- highersuicideattemptrisk[15,16,37]. Lowerlevels ingnessneedsincreasevulnerabilitytosuicidalbehav- of parental understanding were strongly associated ior[5].Schoolengagementindicatorsshowedmean- withsuicideattempts,withnearlythreefoldincreased ingful associations with suicide attempts. Frequent oddsamongthosereportingrareornoparentalun- truancywasassociatedwithhighersuicideattempt derstanding.PreviousGSHS-basedanalysesidentify risk,suggestingthatpersistentschooldisengagement parental support as one of the most important pro- mayweakeninstitutionalattachmentandreduceac- tectivefactorsagainstadolescentsuicideattemptsbe- cesstosupportiveenvironments.Multi-countryGSHS causeitpromotesemotionalsecurity,copingcapac- evidenceindicatesthattruancyislinkedtohighersui- Predictor Category OR 95%CI p Truancy(missedclasseswithoutpermission) (Reference:0days) 1–2days 1.11 [0.93,1.33] 0.255 3–5days 1.02 [0.81,1.25] 0.896 6+days 1.33* [1.11,1.59] 0.002 Physicalactivity(past7days) (Reference:7days) 5–6days 0.72* [0.61,0.84] <.001 3–4days 0.69* [0.60,0.79] <.001 1–2days 0.73* [0.63,0.82] <.001 0days 1.10 [0.94,1.29] 0.218 PEattendancefrequency (Reference:4+days/week) 3days/week 1.10 [0.97,1.25] 0.144 2days/week 1.13* [1.01,1.26] 0.035 1day/week 1.21* [1.10,1.33] <.001 Didnotattend 1.14 [0.96,1.34] 0.136 Note.*p<.05.OR=adjustedoddsratio;CI=confidenceinterval.
Predictor Category OR 95%CI p Dailysedentarytime(outsideschoolwork/sleep) (Reference:<1hour) 1–2hours 1.08 [0.95,1.23] 0.214 3–4hours 1.21* [1.07,1.37] 0.002 5–6hours 1.68* [1.48,1.91] <.001 7+hours 1.62* [1.41,1.86] <.001 Averagenightlysleepduration (Reference:9+hours) 8hours 1.23* [1.08,1.41] 0.003 7hours 1.01 [0.87,1.16] 0.931 5–6hours 1.48* [1.29,1.69] <.001 <5hours 1.93* [1.67,2.23] <.001 Note.*p<.05.OR=adjustedoddsratio;CI=confidenceinterval. cide attempt risk due to reduced adult supervision thatexcessivesedentarytimecontributestopsycho- andincreasedexposuretopsychosocialstressors[20, logicaldistressthroughreducedphysicalactivity,im- 24]. InthePhilippinecontext,thesevereshortageof pairedemotionalregulation,andincreasedsocialiso- school-basedguidancecounselorshasbeenidentified lation[1,23].Shortersleepdurationwasstronglyasso- asastructuralfactorthatconstrainsschools’capac- ciatedwithsuicideattempts;umbrella-levelevidence ity to identify and support students at risk [32]. In identifiessleepdisturbanceasoneofthemostconsis- contrast,physicalactivitydemonstratedaprotective tentlifestylepredictorsofsuicidalbehaviorbecauseit relationship,consistentwithliteratureshowingthat disruptsemotionalregulation,cognitivefunctioning, regularphysicalactivitycontributestoemotionalreg- andstressresponsesystems[23].Thesefindingsare ulation, stress reduction, and improved psychologi- consistentwiththestress–diathesismodel,whichsug- calwell-being[1].ParticipationinPhysicalEducation geststhatphysiologicalstressorssuchassleepdepri- wasalsoassociatedwithlowersuicideattemptrisk, vationcanweakenemotionalresilienceandincrease supportingschool-basedpreventionframeworksem- vulnerabilitytosuicidalbehavior[6].Victimizationex- phasizingthatstructuredschoolactivitiesstrengthen perienceswerestronglyandindependentlyassociated institutionalconnectednessandpromotesupportive withsuicideattempts. Repeatedinjuryduetobully- peerinteractions[1,21,17].Lifestyleregulationfactors ing,bullyingonschoolproperty,andcyberbullyingall alsocontributedsubstantiallytosuicideattemptrisk. showedsignificantassociations.Cross-nationalanal- Prolonged sedentary behavior was associated with ysesacrossbothEuropeanandLMICcontextsconsis- increasedsuicideattempts,consistentwithevidence tentlyidentifybullyingvictimizationasarobustpredic- Predictor Category OR 95%CI p Seriousinjuryduetobullying(past12months) (Reference:0times) 1time 1.02 [0.86,1.19] 0.805 2–3times 1.04 [0.86,1.26] 0.659 4+times 1.42* [1.19,1.69] <.001 Bulliedonschoolproperty(past12months) (Reference:0times) 1–2times 1.20* [1.01,1.41] 0.035 3–5times 1.09 [0.90,1.32] 0.376 6+times 1.70* [1.43,2.02] <.001 Cyberbullying(past12months) (Reference:0times) 1–2times 1.23* [1.03,1.47] 0.024 3–5times 1.10 [0.89,1.36] 0.389 6+times 2.19* [1.84,2.62] <.001 Note.*p<.05.OR=adjustedoddsratio;CI=confidenceinterval. torofsuicidalbehavior[38,27].Cyberbullyingdemon- Several limitations should be acknowledged. First, stratedthestrongestvictimizationassociation,reflect- thestudyreliedonself-reportedsurveydata,which ingthepersistent,pervasive,andsociallyvisiblenature maybesubjecttorecallbiasorsocialdesirabilitybias, of online victimization, which intensifies emotional particularlyforsensitivebehaviorssuchassubstance distressandsocialisolation[23,1]. PhilippineGSHS use,bullying,andsuicideattempts.Second,thecross- analyseslikewisereportthatbullying-relatedvictim- sectionalnatureofthedatalimitstheabilitytoestab- izationsignificantlyincreasessuicideattemptrisk[15, lishcausalrelationships;theresults,therefore,indicate 16,29].ThesefindingssupporttheInterpersonalThe- statisticalassociationsratherthancausaleffects.Third, oryofSuicide, whichsuggeststhatrepeatedexperi- thefindingsreflectschool-basedadolescentpopula- ences of interpersonal harm increase psychological tionsandmaynotfullyrepresentadolescentswhoare vulnerabilityandcontributetosuicidalbehavior[5].In outofschoolorabsentduringsurveyadministration. thePhilippinecontext,thedecriminalizationofsuicide Finally,somevariablesrepresentbroadbehavioralcat- andsuicideattempts—advocatedinrecentWHOpol- egoriesandmaynotcapturethefullcomplexityofun- icyguidance[39]—representsanimportantstructural derlyingpsychologicalorcontextualmechanismsas- enablerforremovingbarrierstohelp-seekingamong sociatedwithsuicideattempts.
victimizedadolescentswhomayfearstigmaorlegal consequences.
6.Recommendations Schoolsshouldstrengthencomprehensivesuicidepre- 5.ContributionsandLimitations ventioneffortsbyimprovingattendancemonitoring, Thisstudyprovidesempiricalevidenceonthemulti- reinforcing evidence-based anti-bullying initiatives, levelpredictorsofsuicideattemptsamongadolescents andmaintainingsafeandsupportivelearningenviron- usingschool-basedhealthsurveillancevariables.The ments [17, 21]. Schools should also encourage stu- findingsdemonstratethatsuicideattemptsareassoci- dentparticipationinPhysicalEducationandstructured atedwithmultipledomainsofadolescentlife,includ- physicalactivities,asengagementinschoolactivities ingsubstanceusebehaviors,relationalconnectedness, and physical activity was associated with lower sui- schoolengagement,lifestyleregulation,andvictimiza- cideattemptrisk.Teachersandguidancecounselors tionexperiences.Byexaminingthesevariablessimulta- shouldstrengthenearlyidentificationmechanismsfor neouslywithinahierarchicallogisticregressionframe- studentswhoexhibitbehavioralriskindicators,par- work,thestudyprovidesacomprehensivepictureof ticularly frequent truancy, repeated bullying victim- howbehavioral,relational,institutional,andlifestyle ization,limitedpeerrelationships,andlowparental factorsintersectinshapingsuicideattemptrisk.The understanding.Regularclassroomobservation,guid- resultsreinforceGSHS-basedevidencethatadolescent anceconsultations,andpeer-supportinitiativesmay suicideriskismultidimensionalandinfluencedbyclus- helpdetectandassiststudentswhomaybeatrisk.At tersofbehavioralandpsychosocialexposures[1,19, thefamilyandpolicylevels,parentsshouldstrengthen 20,23].
opencommunicationandemotionalsupportwithin thehomeenvironment,asparentalunderstandingwas satisfaction. JournalofAdolescentHealth,70(5), stronglyassociatedwithlowersuicideattemptrisk.Ed- 737–745.
ucationandhealthauthoritiesshouldintegratebehav- 12. Sigfusdottir,I.D.,etal.(2018).Suicidalideation ioralriskmonitoringintoschool-basedmentalhealth inadolescence. ActaPsychiatricaScandinavica, programs, particularly targeting substance use, bul- 138(5),440–452.
lying,unhealthysleeppatterns,andexcessiveseden- 13. Hawton,K.,etal.(2022).Suicideriskassessment. tarybehavior. Urgentinvestmentinguidancecoun- TheLancetPsychiatry,9(11),922–928.
selorstaffingisneededtoapproachtherecommended 14. Save the Children Philippines. (2023). Mental 1:500 ratio in Philippine public schools [17]. Future healthofchildren. https://reliefweb.int/ researchersshouldfurtherexaminethesepredictors 15. Chiu,H.,&Vargo,E.J.(2022). Bullyingandsui- usinglongitudinalorcohortdesignstoclarifytemporal cidebehaviors. BMCPsychiatry,22(1),445. relationshipsbetweentheriskdomainsidentifiedin 16. Lagman,J.G.,etal.(2021).Suicideattemptsin thisstudy,andshouldexploreadditionalpsychosocial Filipinoyouths. Cureus,13(9),e18100.
variables—includinggender,sexualorientation,and 17. Alibudbud,R.(2023).Youthsuicideprevention. socioeconomicfactors—thatmayinteractwiththerisk AsianJournalofPsychiatry,83,103538.
18. Hosmer,D.W.,etal.(2013). AppliedLogisticRe- gression.JohnWiley&Sons.
19. Campisi,S.C.,etal.(2020).Suicidalbehaviours