Assessing depressive symptoms in persons who die of suicide in mainland China

Journal of Affective Disorders xx (2006) xxx–xxx

+ MODEL

JAD-03387; No of Pages 10

www.elsevier.com/locate/jad

ARTICLE IN PRESS

Research report

Assessing depressive symptoms in persons who die of suicidein mainland China

Michael Robert Phillips a,⁎, Qijie Shen b, Xiehe Liu c, Sonya Pritzker d,David Streiner e, Ken Conner f, Gonghuan Yang g

a Beijing Suicide Research and Prevention Center, Beijing Hui Long Guan Hospital, Beijing, 100096, People's Republic of Chinab Shenzhen Mental Health Institute, Shenzhen, Chinac West China Medical University, Chengdu, China

d University of California, Los Angeles, USAe Baycrest Centre for Geriatric Care and Department of Psychiatry, University of Toronto, Toronto, Canada

f Department of Psychiatry and Center for the Study and Prevention of Suicide, University of Rochester Medical Center, Rochester, NY, USAg Institute of Basic Medical Sciences, Chinese Academy of Medical Science, Peking Union Medical College, Beijing, China

Received 1 May 2006; received in revised form 31 July 2006; accepted 31 July 2006

Abstract

Background: The potential insensitivity to depression of translated diagnostic instruments makes it difficult to assess the relationship ofdepressive symptoms to suicide in non-Western cultures.Methods: Addition of culturally sensitive probes and other modifications were made to the depression section of the Chinese version ofthe SCID; the standard SCID probes and the expanded-probes are separately used to assess each symptom of depression, the resultantdiagnoses and the overall severity of depression. This modified SCID was included in the psychological autopsy interviews with familymembers and, separately, close associates of 887 suicides and 721 non-suicidal decedents from 23 regions of mainland China.Results: Compared to the standard interview, the expanded-probemethod increased reported prevalence ofmajor depressive episode amongsuicide decedents from 26.4% (234/887) to 40.2% (357/887) and for other deaths from 1.0% (7/721) to 2.1% (15/701). The additional 131cases identified using the expanded-probemethod had substantial social impairment and a greatly elevated risk of suicide compared to thosewith no depressive symptoms (OR=37.0, 95% CI=17.6–77.6). Inter-observer reliability for major depressive episode between the twoindependent interviewswas greater for the expanded probemethod (ICC=0.77 vs. 0.67,Pb0.001). For both interviewmethods there was astrong dose-response relationship between suicide risk and the number and severity of depressive symptoms.Limitations: This study uses proxy informants to obtain information about the psychological status of deceased subjects; the value of thisexpanded-probe method for the diagnosis of depression in non-Western cultures needs to be confirmed with living subjects.Conclusions: Adding culture-appropriate probes about depressive symptoms to standardized diagnostic instruments identifies manyChinese subjects with unrecognized depression. Dimensional measures of depressive symptoms are more powerful predictors of suiciderisk than categorical diagnoses.© 2006 Elsevier B.V. All rights reserved.

Keywords: Depression; Suicide; Diagnostic instruments; Psychological autopsy; China

⁎ Corresponding author. Tel.: +86 10 6271 2471; fax: +86 10 8295 1150.E-mail address: [email protected] (M.R. Phillips).

0165-0327/$ - see front matter © 2006 Elsevier B.V. All rights reserved.doi:10.1016/j.jad.2006.07.020

Please cite this article as: Michael Robert Phillips et al., Assessing depressive symptoms in persons who die of suicide in mainland China,Journal of Affective Disorders (2006), doi:10.1016/j.jad.2006.07.020.

mailto:[email protected]

http://dx.doi.org/10.1016/j.jad.2006.07.020


2 M.R. Phillips et al. / Journal of Affective Disorders xx (2006) xxx–xxx

ARTICLE IN PRESS

1. Introduction

Several reports suggest that rates of depression inChinese populations are lower than those in the west(Parker et al., 2001b). Large epidemiological studiesusing DSM-III criteria and a similar method to the ECAstudies found lifetime prevalences of major depressivedisorder of 1.9% in Taiwan in 1985 (Yeh et al., 1985),and 1.3% in Hong Kong in 1993 (Chen et al., 1993)versus 5.9% in the five ECA sites in the mid-1980s(Robins et al., 1991). A 2002–2003 study (The WHOWorld Mental Health Survey Consortium, 2004) usingDSM-IV criteria reported 12-month prevalences formood disorders of 2.5% in Beijing and 1.7% in Shanghaiversus 9.6% in the United States.

These differences may reflect an actual lowerprevalence of depression in Chinese individuals, butsuch a conclusion would be inconsistent with therelatively high rates of suicide in China (Phillips et al.,2002a) and with other studies in Chinese populations thatreport rates of depressive disorders similar to thosereported in the west (Cheng, 1988; Shi et al., 2005), so avariety of methodological explanations need to beconsidered.

Linguistic and cultural factors are two potentialcontributors to the lower reported rates. Chinese respon-dents may be less willing to report psychological symp-toms (Kleinman, 1986), more likely to somaticizesymptoms (Kleinman, 1986; Parker et al., 2001a,b) ordescribe depressive symptoms in different ways thantheir western counterparts. In mainland China the greateracceptability of the ‘neurasthenia’ label may alsodecrease reports of non-neurasthenic depressive symp-toms (Lee, 1998). Overall, these factors could decreasethe sensitivity of standardized western diagnostic instru-ments to depression in Chinese individuals.

Given the close relationship of depression and suicide,failure to address the potential insensitivity to depressionof translated diagnostic instruments could seriouslyundermine the validity of studies aimed at identifyingrisk factors for suicide, so when designing the nationalcase-control psychological autopsy study of suicide inChina (Phillips et al., 2002b; Phillips and Yang, 2004;Yang et al., 2005) we made specific revisions in thediagnostic procedure aimed at improving its ability todetect depression. We used western diagnostic criteria fordepression (fourth edition of the Diagnostic andStatistical Manual of Mental Disorders, AmericanPsychiatric Association, 1994) (DSM-IV) and a westernstructured questionnaire (the Structured Clinical Inter-view for DSM-IV, First et al., 1996) (SCID), but addedculture-sensitive probes for the symptoms of depression

Please cite this article as: Michael Robert Phillips et al., Assessing depreJournal of Affective Disorders (2006), doi:10.1016/j.jad.2006.07.020.

and made other modifications to the depression section ofthe interview. This strategy made it possible to comparethe prevalence of depression using the expanded probesversus the standard SCID probes and to compare thecharacteristics, clinical severity, and risk of suicide amongpersons identified as depressed using the two methods.

We hypothesized that the expanded-probe methodwould identify a substantial number of depressedsubjects not identified by the standard SCID interview,that these additional cases would have a much higherrisk for suicide than decedents without depressivesymptoms, and that dimensional measures of theseverity of depression using either method would bebetter predictors of suicide risk than the dichotomousdiagnostic category.

2. Methods

2.1. Participants

The national psychological autopsy study is describedin detail in previous reports (Phillips et al., 2002b;Phillips and Yang, 2004; Yang et al., 2005). Conductedin 23 geographically representative sites selected fromthe 145 sites included in China's National DiseaseSurveillance Points network, the study considered thefour recorded causes of death most likely to includesuicides: suicide, non-suicidal injuries, undeterminedinjuries and mental illness. Deaths attributed to thesecauses in persons over 10 years of age that occurred in 3of the sites from 1August 1995 to 31 August 2000 and inthe other 20 sites from 1 January 1997 to 31 August 2000were reported to the research group at 3-month intervals.All deaths with undetermined injury or mental illness asthe recorded cause were selected for detailed investiga-tion. Deaths with suicide or non-suicidal injury as therecorded cause occurring prior to March 1, 1998 wereconsecutively sampled; starting in March 1998, if morethan 5 suicides or 5 non-suicidal injury deaths werereported in any 3-month period, 5 cases of each type ofdeath were randomly selected for detailed investigation,if fewer than 5 cases of either type were reported, all wereselected and additional cases were consecutively sam-pled from the death registries of adjacent counties ortownships.

Trained researchers visited each site at regularintervals to administer the comprehensive psychologicalautopsy interview schedule – which takes 2–3 h tocomplete – to family members and, separately, to closeassociates of the 1807 deceased subjects selected fordetailed investigation. Appropriate respondents couldnot be located for 67 subjects, identified respondents for

ssive symptoms in persons who die of suicide in mainland China,


3M.R. Phillips et al. / Journal of Affective Disorders xx (2006) xxx–xxx

ARTICLE IN PRESS

65 subjects refused to participate, the data for 22subjects was lost in the mail, the psychiatric interviewfor 4 subjects was incomplete, and data for 41 subjectsunder 16 years of age at the time of death were excludedfrom this analysis. Based on the final classification ofthe cause of death, the remaining 1608 subjects included887 suicides (cases) and 721 non-suicidal deaths(controls) attributed to injury (n=687) or mental illness(n=34). Compared to controls suicide decedents wereolder [46.1(19.3) vs. 44.2 (18.3); t=1.97, df=1606,P=0.047], less likely to be male [50.7% vs. 75.3%;X2 =101.73, df=1, Pb0.001] and more likely to live in arural village [78.9% vs. 70.7%; X2 =14.3, df=1,Pb0.001].

Two valid interviews were conducted for 1493subjects and one valid interview for 115 subjects.Among the 3101 interviews, a single proxy informantparticipated in 2404 (78%) interviews and multipleinformants in 697 (22%); 1541 (50%) interviews hadmale informants, 1115 (36%) female informants and 445(14%) both male and female informants; the mean (sd)age of informants was 46 (13); 2482 (78%) interviewswere conducted by psychiatrists, 354 (11%) bypsychiatric nurses and 319 (10%) by public healthdoctors. Median (IQR) time from death to interview was11.0 (7.8–14.2) months.

2.2. Revisions to the depression section of the SCID

The SCID was revised for use with proxy informantsand the section on depression was translated and backtranslated several times to ensure parallelism with thestandard probes used in the original English version.With the exception of the item on agitation (which wasconsidered appropriate for China without revision), 1–3culture-appropriate probes were added for each symp-tom of depression and asked after the standard SCIDprobe(s) unless responses to the standard probesindicated that the symptom was definitely present forat least 14 days prior to death. The added probes wereoriginally developed by two senior psychiatrists (QSand XL) familiar with depression as it is experiencedand expressed by Chinese individuals; they weredesigned to evoke the same symptom as the originalprobe but are phrased in language that is morecommonly used to describe the experience of thesymptom in China. For example, the standard probefor the depressive affect item asks about feeling‘depressed’ (a translated word which is not a frequentlyused term in everyday Chinese) and feeling ‘down’,while the expanded probes ask about the tendency toweep frequently for no apparent reason, the feeling of


‘men men bu le’ (a common term that essentially meansbeing in low spirits), being gloomy and pessimistic, andso forth. These expanded probes were refined duringtwo years of extensive field trials; the present study usesthe fifth version of the interview. The inter-raterreliability for affective disorders of the revised instru-ment among 16 coders who independently evaluated 37tape-recorded interviews was good (ICC=0.87).

We also made several other modifications to thedepression section of the SCID. (1) For each symptomthat was definitely present or considered ‘subthreshold’the number of days the symptom was continuouslypresent prior to death was recorded; (2) all ninesymptoms of depression were assessed even if depres-sive affect or loss of interest were absent; (3) the socialand psychological effect of the depressive symptoms inthe month prior to death was assessed by asking abouthow distressed the individual was by the symptoms andabout the effect the symptoms had on the individual'swork, daily activities, spirits, social interactions andself-care (each of the six items was scored 0 to 3, none tosevere); (4) the overall effect of the deceased'sdepressive symptoms on family members in the monthbefore death was assessed on a 0 to 3 (none to severe)scale.

2.3. Measures

Eight depression-related variables were computed foreach interview.

(1) Presence of each of the 9 symptoms of depressionusing standard and expanded probes. A symptomwas considered ‘present’ using standard probes ifresponses to the standard probe indicated it wasdefinitely present for at least 14 days prior todeath; it is considered present using expandedprobes if responses to standard or expandedprobes indicated it was definitely present for atleast 14 days. (For the ‘thoughts of death’ item,occurrence of a suicide attempt or a specificsuicidal plan anytime in the 14 days prior to deathis sufficient.)

(2) Number of standard-probe and expanded probesymptoms present (range 0-9).

(3) Diagnostic result using standard and expandedprobes. Symptom patterns were classified intofour ranked categories: major depressive episode(N5 symptoms including depressive affect or lossof interest), minor depressive episode (2–4symptoms including depressive affect or loss ofinterest), subthreshold symptoms (any of the 9




ARTICLE IN PRESS

symptoms coded as ‘definite’ or ‘subthreshold’ forany duration in the 14 days prior to death inpersons who do not meet criteria for major orminor depressive episode), and no symptoms.

(4) Severity of depressive symptoms in the monthbefore death using standard and expanded probes.The score for each probe is the product of theseverity (not present=0; subthreshold=1, definite-ly present=2) and the number of days the symptomwas present in the month prior to death. Thestandard-probe symptom score is the maximumscore of standard probes for the symptom (somesymptoms have multiple standard probes), and theexpanded-probe symptom score is the maximumscore of all the probes for the symptom. (For thethoughts of death symptom, occurrence of a suicideattempt or a specific suicidal plan at anytime in themonth prior to death was assigned the maximumscore of 60.) The overall standard-probe andexpanded-probe scores are the sum of the 9symptom scores (range 0–540); this is divided by5.4 to generate scores with a range of 0–100 (leastto greatest severity). The advantage of this measureover the symptom count measure is that itquantifies subthreshold depressive symptoms andsymptoms that have lasted for less than 14 days.

(5) Effect of depressive symptoms on the individual inthe month before death. The sum of the six itemsis divided by 0.18 to generate a score with a rangeof 0–100 (least to most severe effect).

(6) Effect of the deceased's depressive symptoms onfamily members in the month before death. Theoriginal item score of 0–3 (least to most severeeffect) is employed.

(7) Quality of life in the month prior to death.Respondents rate six characteristics of the deceased(physical health, psychological health, economiccircumstances, work, family relationships, andrelationships with non-family associates) on ascale of 1 (very poor) to 5 (excellent) and the sumof the six scores is converted to a scale of 0–100.

(8) Chronic stress score in the last year. Derived froma 60-item, interviewer-completed life event scale(constructed and pre-tested specifically for thisproject), chronic stress was the combined sum ofthe product of the duration (in months) andseverity of the psychological effect (on a scale of0–4) of each negative life event.

When combing the two interviews to determine theresults for a specific subject the mean value was used forcontinuous measures, but categorical variables were


coded ‘up’: that is, if the results for the two interviewsvaried for a specific depressive symptom the symptomwas considered present, and if the diagnostic result variedthe more ‘severe’ diagnosis was assigned to the subject.

After complete description of the study to therespondents, verbal informed consent was obtained.The study was approved by the institutional reviewboards of the Beijing Hui Long Guan Hospital and theChinese Academy of Preventive Medicine.

2.4. Statistical methods

SPSS (version 11.0) software was used to conductthe analysis. Concordance of results from the family-informant interview and the independent non-familyinformant interview for both standard-probes andexpanded-probes were assessed using intraclass corre-lation coefficients (ICC) computed using a two-waymixed model for the average of the two ratings. Oddsratios (OR) of suicide (versus other deaths) wereadjusted for age (6 categories), gender, location ofresidence (villages vs. towns and cities) and the age bygender interaction term. Linearity of suicide risk forranked variables was assessed using orthogonal poly-nomial contrasts in the logistic regression equations(Nagelkerke, 1991). The proportion of explainedvariation in the logistic models was estimated usingthe Nagelkerke R2 statistic. Comparison of the ICCs,ORs and Nagelkerke R2 values for the standard-probeand expanded-probe methods was based on theconfidence interval of the difference in the parametervalues estimated using a bootstrap procedure thatsampled 1000 times with replacement. The character-istics of the four diagnostic subgroups of persons whodied by suicide were compared using chi square tests(for categorical variables) or Kruskal-Wallis tests (forcontinuous variables); if significantly different, multiplecomparisons methods assessed differences betweengroups — for categorical variables, a Tukey-typemethod based on an arcsin transformation of the originalproportions is used and for continuous variables a non-parametric method for subgroups with unequal samplesizes that compares mean ranks between groups andadjusts for tied ranks is used (Zar, 1999).

3. Results

Table 1 shows that with the exception of the poorconcentration item when using the standard-probemethod (ICC=0.47), agreement between the twoindependent interviews with different informants forthe nine depressive symptoms using both the standard-



Table 1Concordance of results between two independent interviews (with different proxy informants) when using standard-probe and expanded-probeversions of the nine SCID depression items for 1493 deceased persons in mainland China whose recorded causes of death were suicide, injury andmental illness

Using standard probes Using expanded probes Two-tailedP-valuea

ICC ICC

Presence of specific symptomsb

Depressed mood 0.71 0.78 b0.001Diminished Interest 0.67 0.74 0.004Weight change 0.66 0.65 N0.05Sleep disorder 0.67 0.70 0.006Agitationc 0.66 0.66 1.000Fatigue 0.60 0.68 b0.001Worthlessness 0.64 0.67 N0.05Poor concentration 0.47 0.67 b0.001Thoughts of death 0.73 0.77 0.036

Results of interviewPresence of major depressive episode at time of death 0.67 0.77 b0.001Number of symptoms present in last 2 weeks (range 0–9) 0.79 0.82 0.002Severity of depressive symptoms in last month 0.80 0.82 0.004

ICC: intraclass correlation coefficient.a Based on the confidence interval of the difference score between the two ICC values estimated using a bootstrap procedure (sampling 1000

times with replacement).b Symptom considered ‘present’ if respondent reports that it was definitely present for at least 14 days prior to death.c No expanded probes were used for the item on agitation.


ARTICLE IN PRESS

probe and expanded-probe method was good (ICCs0.61–0.78). Inter-observer reliability was significantlybetter when using the expanded-probe method for thediagnosis of major depressive episode, the symptomcount measure, the symptom severity measure, and forsix of the nine symptoms. Concordance between the twointerviews of other measures employed in the analysiswas also good: psychosocial effect of depression on theindividual (ICC=0.80), effect of decedents' depressivesymptoms on family members (0.74); quality of lifescore (0.82); and chronic stress score (0.77).

Table 2 shows that use of the expanded probes lead to a50% increase in the reported prevalence of majordepressive episode in persons who died by suicide (from26.4% to 40.2%) and a 100% increase in persons who diedof other causes (from 1.0% to 2.1%). The risk of suicidefor persons with major depressive episode (versus allothers) was similar using the two diagnostic methods-adjusted OR (95% CI) for the standard-probe method was39.2 (18.2–84.4) versus 33.9 (19.8–58.1) for the expand-ed-probe method (PN0.05), but the expanded-probemethod explained more of the variance (R2=0.39 vs.0.31, Pb0.001).

Among the 131 subjects who meet criteria for majordepressive episode when using the expanded-probemethod but not the standard-probe method, the mean(sd) detriment in psychosocial functioning in the month


prior to death attributed to depressive symptoms was43.0% (20.4%) and the mean quality of life score (on a0–100 scale) was only 51.5 (10.6), so they weresubstantially impaired by their symptoms. Compared tothe 928 subjects with no depressive symptoms, theadjusted OR (95% CI) for suicide in these 131 subjectswas 37.0 (17.6–77.6). These findings help to confirm thepredictive validity of the modified diagnostic instrument.

With the exception of the sleep disorder item(Pb0.001) the risk of suicide associated with specificdepressive symptoms was not significantly differentbetween the two methods; for both methods ‘thoughts ofdeath’ and ‘worthlessness’ were the symptoms moststrongly associated with suicide risk (Table 2). For boththe standard-probe and expanded-probe methods the riskof suicide in the four diagnostic categories (no symptoms,subthreshold symptoms, minor depressive episode, majordepressive episode) showed a strong linear trend and nosignificant non-linear components. This ranked 4-leveldiagnostic categorization explained more of the variancethan the dichotomous categorization of a major depressiveepisode (for the standard-probe categories R2=0.49 vs.0.31, Pb0.001; and for expanded-probe categoriesR2=0.51 vs. 0.39, Pb0.001). There is no difference inthe prevalence of other (non-depressive) mental disordersbetween suicides with no depressive symptoms (usingexpanded probes) and those with subthreshold symptoms



Table 2Risk of suicide in persons who experienced symptoms of depression in the 2 weeks before death as reported by proxy informants administered arevised version of SCID that includes both standard probes and expanded probes for depressive symptoms

Symptom/diagnosis Symptom/diagnosis presenta using standard probes Symptom/diagnosis presenta using expanded probes

Suicides[N=887]

Other deaths[N=721]

Adjusted ORb

(95% CI)Suicides[N=887]

Other deaths[N=721]

Adjusted ORb

(95% CI)

n (%) n (%) n (%) n (%)

Separate symptomsDepressed mood 323 (36.4) 17 (2.4) 25.3 (15.2–42.1) 444 (50.1) 29 (4.0) 26.6 (17.7–39.9)Diminished interest 235 (26.5) 12 (1.7) 22.9 (12.6–41.7) 376 (42.4) 21 (2.9) 25.6 (16.1–40.7)Weight change 223 (25.1) 21 (2.9) 11.7 (7.3–18.7) 258 (29.1) 24 (3.3) 12.5 (8.0–19.5)Sleep disorder 319 (36.0) 27 (3.7) 14.4 (9.5–21.8) 334 (37.7) 27 (3.7) 15.7 (10.3–23.8)Agitationc 359 (40.5) 30 (4.2) 16.6 (11.2–24.8) 359 (40.5) 30 (4.2) 16.6 (11.1–24.8)Fatigue 259 (29.2) 27 (3.7) 11.5 (7.5–17.5) 328 (37.0) 30 (4.2) 14.4 (9.6–21.4)Worthlessness 202 (22.8) 6 (0.8) 40.3 (17.6–92.0) 295 (33.3) 9 (1.2) 43.8 (22.2–86.6)Poor concentration 109 (12.3) 10 (1.4) 11.7 (6.0–22.9) 258 (29.1) 17 (2.4) 17.6 (10.6–29.4)Thoughts of death 353 (39.8) 7 (1.0) 67.9 (31.7–145.4) 433 (48.8) 12 (1.7) 56.9 (31.5–103.0)

Results of interviewd

No symptoms 306 (34.5) 649 (90.0) 1.0 286 (32.2) 642 (89.0) 1.0Subthreshold symptoms 232 (26.2) 53 (7.4) 9.6 (6.8–13.6) 148 (16.7) 51 (7.1) 6.5 (4.5–9.5)Minor depressive episode 115 (13.0) 12 (1.7) 21.4 (11.4–40.0) 96 (10.8) 13 (1.8) 18.4 (9.9–34.0)Major depressive episode 234 (26.4) 7 (1.0) 79.3 (36.6–172.0) 357 (40.2) 15 (2.1) 59.1 (34.2–102.2)a Symptom considered ‘present’ if respondent reports it was definitely present for at least 14 days prior to death in either of the two interviews

conducted for each case.b Odds ratios adjusted for age (6 categories), gender, location of residence (villages vs. towns and cities) and the age by gender interaction in a

logistic regression equation. With the exception of the sleep disorder item (Pb0.001), there were no significant differences in the standard-probe andexpanded-probe odds ratios.c No expanded probes were used for the item on agitation.d Major and minor depressive episodes defined according to DSM IV criteria; ‘subthreshold symptoms’ is coded when interview results do not

meet criteria of major or minor depressive episode but indicate that the deceased experienced any of the nine symptoms of depression at any time inthe two weeks before death (not necessarily for the full 14 days). If results of the two interviews (with different proxy informants) are different, thecase is classified as the more severe category. Nagelkerke R2 for the logistic models of four standard-probe and four expanded-probe diagnosticcategories were 0.49 and 0.51, respectively (P=0.006).


ARTICLE IN PRESS

or minor depressive disorder [38% (109/286) vs. 40% (98/244); X 2=0.23, df=1, P=0.629], so the presence of otherpsychiatric diagnoses does not explain the substantialincrease in suicide risk among persons with subthresholddepressive symptoms or minor depressive episode.

Considering the 887 suicides and 701 non-suicidedeaths separately, the number, severity and psychosocialeffect of depressive symptoms reported (by eithermethod) did not differ significantly by time intervalsince death, type of interviewer (psychiatrist, psychiatricnurse or public health doctor), between subjectsconsecutively or randomly sampled, or between subjectsfor whom one or two proxy interviews were conducted.But male proxy respondents for suicides reported moredepressive symptoms and a greater psychosocial effect ofdepressive symptoms than female proxy informants, andmore depressive symptoms were reported for non-suicidedecedents if multiple proxy respondents participated inthe interview than if a single proxy respondentparticipated (data available on request).


Table 3 compares characteristics of four subgroups ofsuicides: 286 with no depressive symptoms, 244 withsubthreshold symptoms or a minor depressive disorder,123 who meet criteria of major depressive episode whenusing expanded probes but not when using standardprobes, and 234 who met criteria of major depressiveepisode using standard probes. Persons with standard-probe major depressive episode were older, more likelyto have made a prior suicide attempt, more likely to havereceived psychological treatment, and most severelyaffected by their depressive symptoms. In cases thatonly met major depressive episode criteria when usingexpanded probes the depressive symptoms had a lesspronounced psychosocial effect on the individual andhis or her family, were less likely to have lead topsychological treatment, and were more likely to beaccompanied by an acute precipitating life event in theweek before death. Persons who die of suicide in theabsence of depressive symptoms are younger, are lesslikely to have made a prior suicide attempt, have lower



Table 3Comparison of different subgroups of 887 persons who died by suicide from mainland Chinaa

Characteristic (A) (B) (C) (D) Chisquare

df Pb Multiplecomparisonsc

No depressivesymptoms[N=286]

Sub-thresholdsymptoms or minordepressive episode[N=244]

Expanded-probemajor depressiveepisode [N=123]

Standard-probemajor depressiveepisode [N=234]

n (%) n (%) n (%) n (%)Male 130 (45.5) 125 (51.2) 66 (53.7) 129 (55.1) 5.44 3 0.142 all NSMarried at time of deathd 196 (68.8) 155 (63.5) 79 (64.2) 161 (68.8) 2.46 3 0.482 all NSLived in rural village 226 (79.0) 197 (80.7) 103 (83.7) 174 (74.4) 5.13 3 0.163 all NSPrecipitating event in week

prior to deathd120 (42.0) 77 (31.7) 38 (30.9) 38 (16.2) 40.06 3 b0.001 A, BND***;

CND**Previous suicide attempt 49 (17.1) 74 (30.3) 30 (24.4) 86 (36.8) 27.25 3 b0.001 DNA***;

BNA**Any prior psychological

treatment47 (16.4) 35 (14.3) 19 (15.4) 64 (27.4) 16.45 3 b0.001 DNB**;

DNA, C*Used psychiatric

medications in monthbefore death

28 (9.8) 19 (7.8) 9 (7.3) 36 (15.4) 9.42 3 0.024 DNB*

Had prior episode(s)of depression

–e –e 25 (20.3) 56 (23.9) 0.60 1 0.439 all NS

M (IQR) M (IQR) M (IQR) M (IQR)Age 36 (27–59) 40 (29–63) 46 (32–67) 49 (33–63) 15.16 3 0.002 DNA**;

CNA*Years of schoolingd 5 (0–7) 5 (0–8) 5 (0–8) 5 (0–8) 3.16 3 0.368 all NSMonthly per capita family

income ($US)d18 (9–36) 15 (8–24) 15 (7–24) 18 (10–36) 9.40 3 0.024 all NS

Quality of life in lastmonth (0–100)d,f

58 (46–67) 52 (44–61) 50 (44–58) 46 (38–55) 73.88 3 b0.001 ANB, C,D***;BND***;CND*

Chronic stress scoreover last yeard,f

36 (12–72) 54 (25–99) 54 (31–88) 72 (40–121) 57.82 3 b0.001 D, BNA***;CNA**;DNB*

Effect of depression inlast month (0–100)f

–e 17 (3–39) 42 (28–56) 58 (42–78) 194.38 2 b0.001 DNCNB***

Duration of episode attime of death (days)

–e 14 (2–59) 53 (30–120) 69 (32–180) 115.92 2 b0.001 D, CNB***

Effect of depression onfamily (0–3)f

–e 0.5 (0–1.0) 1.0 (0.5–2.0) 2.0 (1.0–2.0) 134.25 2 b0.001 DNCNB***

M=median, IQR=interquartile range; NS=not significant,*Pb0.05, **Pb0.01, ***Pb0.001.a ‘Standard-probe major depressive episode’ are cases that meet DSM-IV criteria using the results of standard probes in either of the two SCID

interviews (with different proxy informants); ‘expanded-probe major depressive episode’ are those that meet DSM-IV criteria using expandedprobes but do not meet criteria using standard probes in either interview; ‘subthreshold symptoms or minor depressive episode’ are cases that do notmeet criteria of major depressive episode in which responses to either standard or expanded probes in either of the two interviews indicate that thevictim experienced any of the nine symptoms of depression at any time in the 2 weeks before death.b P-value for categorical variables are based on Chi-square tests and those for ranked or continuous variables are based on Kruskal-Wallis tests.c Comparison between groups for categorical variables is conducted using a Tukey-type multiple comparison method based on an arcsin

transformation of the original proportions; comparison between groups for ranked variables uses a non-parametric multiple comparison method forsubgroups with unequal sample sizes that compares mean ranks between groups and adjusts for tied ranks [16].d Missing data: 285 deceased subjects with ‘no depressive symptoms’ had data on marital status and years of schooling and 282 had data on

income; 243 subjects with ‘subthreshold symptoms or minor depressive episode’ had data on precipitating life events, income, quality of life andchronic stress; and 120 subjects with ‘expanded-probe major depressive episode’ had data on income.e This data was not collected for this subgroup of subjects.f Computation of this variable is described in the methods section.


ARTICLE IN PRESS

Please cite this article as: Michael Robert Phillips et al., Assessing depressive symptoms in persons who die of suicide in mainland China,Journal of Affective Disorders (2006), doi:10.1016/j.jad.2006.07.020.


Table 4Risk of suicide based on quintiles of the mean number of depressive symptoms experienced in the 2 weeks prior to death and quintiles of the meanseverity of depressive symptoms in the month prior to death, as reported by proxy informants administered a revised version of SCID that includesboth standard probes and expanded probes for depressive symptomsa

Number and severity of symptoms using standard probes Number and severity of symptoms using expanded probes

Rangeb Suicides[N=887]

Other deaths[N=721]

Adjusted ORc

(95% CI)Rangeb Suicides

[N=887]Other deaths[N=721]

Adjusted ORc

(95% CI)

n (%) n (%) n (%) n (%)

Mean number of symptomsd (range 0–9) Mean number of symptomsd (range 0–9)0 352 (39.7) 663 (92.0) 1.0 0 329 (37.1) 659 (91.4) 1.00.5–1.0 104 (11.7) 28 (3.9) 7.0 (4.4–11.1) 0.5–1.5 115 (13.0) 32 (4.4) 6.8 (4.4–10.6)1.5–2.5 157 (17.7) 20 (2.8) 15.4 (9.4–25.4) 2.0–3.0 113 (12.7) 17 (2.4) 14.6 (8.5–25.2)3.0–4.0 114 (12.9) 7 (1.0) 32.6 (14.9–71.6) 3.5–5.5 157 (17.7) 10 (1.4) 34.6 (17.8–67.3)≥4.5 160 (18.0) 3 (0.4) 120.8 (37.9–384.9) ≥6.0 173 (19.5) 3 (0.4) 127.7 (40.1–406.1)

Mean severity of symptomsd (range 0–100) Mean severity of symptomsd (range 0–100)0 306 (34.5) 649 (90.0) 1.0 0 286 (32.2) 642 (89.0) 1.00.1–10.8 115 (13.0) 34 (4.7) 7.3 (4.8–11.2) 0.1–13.4 124 (14.0) 46 (6.4) 6.1 (4.1–9.0)10.9–25.1 150 (16.9) 25 (3.5) 13.3 (8.4–21.1) 13.5–31.8 136 (15.3) 18 (2.5) 18.5 (10.9–31.4)25.2–46.3 156 (17.6) 10 (1.4) 34.2 (17.5–66.6) 31.9–56.6 169 (19.1) 12 (1.7) 33.6 (18.2–62.3)≥46.4 160 (18.0) 3 (0.4) 135.0 (42.3–430.5) ≥56.1 172 (19.4) 3 (0.4) 146.3 (45.9–465.9)

a The ‘mean number of symptoms’ is the mean number of depressive symptoms that were rated as ‘definitely present’ for at least 14 days prior todeath in the two independent interviews. The ‘mean severity of symptoms’ is the mean severity of depressive symptoms in the two interviews (themethod of computing severity of depressive symptoms is described in Methods).b The cut-off points for the ranges were those that resulted in the most even distribution of subjects across the four non-zero groups.c Odds ratios adjusted for age (6 categories), gender, location of residence (villages vs. towns and cities) and the age by gender interaction in a

logistic regression equation. The odds ratios for all four measures (the number and severity of symptoms assessed using standard or expanded probes)have significant linear trends. There are no significant differences between the standard-probe and expanded-probe odds ratios.d Nagelkerke R2 for the logistic models of standard-probe and expanded-probe quintiles of symptom counts were 0.47 and 0.49, respectively

(Pb0.001); R2 for standard-probe and expanded-probe quintiles of symptom severity were 0.49 and 0.51, respectively (P=0.002).


ARTICLE IN PRESS

chronic stress, have a higher quality life, and are morelikely to have experienced a precipitating life event.

The 357 suicide cases that meet criteria of a majordepressive episode using either method included 307(86%) with a final primary diagnosis of majordepressive disorder, 18 with substance dependencedisorders, 14 with psychotic disorders, 9 with organicmental disorders, 7 with bipolar disorders, and 2 withother disorders.

Table 4 shows quintiles of the symptom count measureand the symptom severity measure for both diagnosticmethods; the ORs for suicide in the four sets of quintiles allhave significant linear trends with no significant non-linearcomponents. This is strong evidence for a ‘dose–response’relationship between suicide risk and the number andseverity of depressive symptoms.

4. Discussion

4.1. Inclusion of culture-sensitive probes in standardizeddiagnostic instruments

The Global Burden of Disease study (Murray andLopez, 1996) estimates that by 2020 neuropsychiatric


disorders will account for 15% of the total burden of illnessin the world and that 83% of this burden will be borne bydeveloping countries. To meet this emerging public healthcrisis governments in developing countries need todevelop, test and promulgate comprehensive programs toexpand the availability, range and quality of mental healthservices; increase awareness of the importance of mentalhealth; and change attitudes about care-seeking for mentalhealth problems. The development and assessment of thesenational mental health action plans must be based onreliable and valid estimates of the prevalence and demo-graphic distribution of different mental disorders, so theessential first step is to develop procedures that can accu-rately diagnose these conditions in developing countries.

We are still a long way from this goal. The recent WHOWorld Mental Health Survey (The WHO World MentalHealth Survey Consortium, 2004) found that even whenusing standardized diagnostic criteria and a fully structureddiagnostic instrument that was formally translated andback-translated there was a 12-fold difference in the cross-national prevalence of mood disorders (0.8–9.6%) and a 6-fold difference in the prevalence of any DSM-IV disorder(4.3–26.4%). Some cross-national differences in theprevalence of mental disorders are expected, but not of




ARTICLE IN PRESS

this magnitude. Several methodological problems couldcontribute to these reported differences – differences indiagnostic criteria, in sampling procedures, in responserates, in the selection and training of interviewers, inrespondent's willingness to provide information, and soforth – but one of the most important issues is the potentialinsensitivity of western diagnostic instruments to psycho-logical symptoms in non-western cultures.

The current study adopts a novel approach foraddressing this issue. We retained the core structure anddiagnostic criteria of the SCID interview for DSM-IV, butadded culture-sensitive probes for depressive symptoms.This strategy dramatically increased the identification ofsubjects with depressive symptoms and, somewhatsurprisingly, also resulted in a statistically significantincrease in the inter-observer reliability of the diagnosisof major depressive episode. The inclusion of additionalprobes for symptoms increases respondent burden, but itmakes it possible to reliably identify symptoms even whenthe expression of symptoms in the community isheterogeneous—which is likely to occur in developingcountries such aChinawhere the awareness and experienceof psychological phenomena in urban and rural commu-nities can be quite different.

Are the additional cases identified using the expandedprobes sufficiently similar to cases identified using standardprobes to warrant assigning the same diagnosis? We foundthat ‘new’ cases of major depressive episode identified bythe expanded-probe method are not as severely affected bythe symptoms as standard-probe cases, but they do havesignificant psychosocial impairment (median level of 42 on0–100 scale) and a greatly elevated risk of suicide(OR=37.0), so they probably merit treatment and, thus, adiagnosis of major depressive episode.

Our study employed proxy informants for deceasedsubjects so the utility of the added probes needs to beconfirmed in large epidemiological surveys of communityresidents. We expect that addition of culture-sensitiveprobes would also improve identification of other SCIDdisorders (anxiety disorders, substance abuse disorders,etc.) and that this procedure could be employed with otherstructured or semi-structured diagnostic interviews and inother non-western cultures. Developing and testing suchprobes is a lengthy process, but it is an essential step toimproving the validity of psychiatric epidemiologicalstudies in non-western cultures.

4.2. Dimensional versus categorical relationship ofdepression to suicide risk

China has a unique pattern of suicides (Phillips et al.,2002a; Yang et al., 2005), so it is unclear how gen-


eralizable our results are to other countries; but the studyraises serious concerns about the dimensional versuscategorical relationship of potential risk factors to suicidethat have not been adequately addressed in any country.Our results clearly demonstrate that the risk of suicideincreases linearly with increased severity of depression andthat dimensional measures of depression were betterpredictors of suicide than the dichotomous diagnosticclassification. Our previous reports (Phillips et al., 2002b;Phillips and Yang, 2004) show that this linear relationshipof depressive severity and suicide risk persists even whenseveral other risk factors are simultaneously considered.

Almost all psychological autopsy studies (Cavanaghet al., 2003) rely heavily on dichotomous diagnosticcategories so they underestimate the importance ofsubsyndromal depression and anxiety and fail to accountfor the contribution of personality traits (such as aggressionand impulsiveness) that may be best measured on acontinuum (Krueger, 1999; Conner et al., 2001). Somenon-diagnostic suicide risk factors may also be bestcaptured using continuous measures: our national study(Phillips et al., 2002b) found that chronic stress, acutestress, and quality of life show a strong dose-responserelationship to suicide risk. Thus improved prediction andmanagement of suicidal behavior will require a funda-mental change in the strategy for assessing risk: diagnosticassessments must be complemented with dimensionalmeasures of psychopathology and other risk factors.

Community screening for depression and diagnosticassessment of persons in high-risk groups are cornerstonesof most national and regional suicide prevention efforts(Gaynes et al., 2004), so this debate about the dimensionalversus categorical relationship of depression to suicide riskhas important implications for suicide prevention activi-ties. Our findings – which need to be confirmed usingliving samples – suggest that diagnosis-based depressionscreening programs may misclassify as ‘low-risk’ largenumbers of subjects with subthreshold depressive symp-toms who, nevertheless, have a substantially higher risk ofsuicide than those without any depressive symptoms.Many screening schedules for depression ‘skip out’ if therespondent does not report depressive affect or loss ofinterest; so respondents are not asked about otherdepressive symptoms that may be more predictive ofsuicide such as thoughts of death or worthlessness. And inour study 60% of suicide cases with subthresholddepressive symptoms did not have another psychiatricdiagnosis so they would not have been identified by moreextensive screening programs that consider a range ofmental disorders. Research comparing the predictivepower and cost-effectiveness of dimensional versuscategorical measures of suicide risk both for community-




ARTICLE IN PRESS

based screening programs and for clinical assessmentprocedures of high-risk groups is urgently needed.

Improved screening programs based on dimensionalmeasures of depression will bring increasing numbers ofindividuals with subsyndromal depression to the attentionof mental health professionals. Clinicians will then befaced with the difficult task of balancing the danger ofpathologizing normal sadness and wasting limited mentalhealth resources against the risk of suicide. The clinicaldecision about whether or not to initiate treatment forsubsyndromal depression must be based on non-diagnosticissues related to suicide risk: personality factors, psycho-social stressors and social support networks.

Acknowledgements

The national psychological autopsy study wassupported by grants from the Ford Foundation, theSave the Children Fund, and Befrienders International.Manuscript preparation was supported by NIH grantR25 #MH68564 (Caine, PI). The project was jointlycoordinated by the Beijing Hui Long Guan Hospital andthe Chinese Academy of Preventive Medicine under thedirection of Drs. Michael Phillips and Gonghuan Yang.The authors thank the Provincial Epidemic PreventionCenters in the 16 provinces (Anhui, Fujian, Guangxi,Guizhou, Hebei, Heilongjiang, Henan, Hunan, Jiangxi,Jilin, Shaanxi, Shandong, Shanxi, Sichuan, Qinghai,Zhejiang); the four participating psychiatric centers(Department of Neuropsychiatry, Xijing Hospital, Xian,Shaanxi Province; Jingzhou City Psychiatric Hospital,Hubei Province; Shenyang Mental Health Center,Liaoning Province; and Suzhou Guangji Hospital,Jiangsu Province); and Yali Zhang and Siului Hui fortheir assistance in the statistical analysis.

References

American Psychiatric Association, 1994. Diagnostic and StatisticalManual of Mental Disorders, 3rd edn. APA, Washington, DC.DSM-IV.

Cavanagh, J.T.O., Carson, A.J., Sharpe, M., et al., 2003. Psychologicalautopsy studies of suicide: a systematic review. PsychologicalMedicine 33, 395–405.

Chen, C.N., Wong, J., Lee, N., et al., 1993. The Shatin mental healthsurvey in Hong Kong. Archives of General Psychiatry 50,125–133.

Cheng, T.A., 1988. A community study of minor psychiatric morbidityin Taiwan. Psychological Medicine 18, 953–968.

Conner, K.R., Duberstein, P.R., Conwell, Y., et al., 2001. Psycholog-ical vulnerability to completed suicide: A review of empiricalstudies. Suicide and Life-Threatening Behavior 31, 367–385.


First, M.B., Spitzer, R.L., Gibbon, M., Williams, J.B., 1996. StructuredClinical Interview for DSM-IVAxis I Disorders. Biometrics ResearchDepartment, New York State Psychiatric Institute, New York.

Gaynes, B.N., West, S.L., Ford, C.A., et al., 2004. Screening forsuicide risk in adults: a summary of the evidence for the U.S.Preventive Services Task Force. Annals of Internal Medicine 140,822–835.

Kleinman, A., 1986. Social Origins of Distress and Disease:Neurasthenia, Depression, and Pain in Modern China. YaleUniversity Press, New Haven.

Krueger, R., 1999. The structure of common mental disorders.Archives of General Psychiatry 56, 921–926.

Lee, S., 1998. Estranged bodies, simulated harmony, and misplacedcultures: Neurasthenia in contemporary Chinese society. Psycho-somatic Medicine 68, 448–457.

Murray, C.J.L., Lopez, A.D., 1996. Global Health Statistics: ACompendium of Incidence, Prevalence, and Mortality Estimatesfor Over 200 Conditions. Harvard University Press, Cambridge,USA.

Nagelkerke, N.J.D., 1991. A note on general definition of thecoefficient or determination. Biometrika 78, 691–692.

Parker, G., Cheah, Y.C., Roy, K., 2001a. Do the Chinese somaticizedepression? A cross-cultural study. Social Psychiatry and Psychi-atric Epidemiology 36, 287–293.

Parker, G., Gladstone, G., Chee, K.T., 2001b. Depression in theplanet's largest ethnic group: The Chinese. American Journal ofPsychiatry 158, 857–864.

Phillips, M.R., Yang, G.H., 2004. Suicide and attempted suicide inChina, 1990–2002. Morbidity and Mortality Weekly Report 53,481–484.

Phillips, M.R., Li, X.Y., Zhang, Y.P., 2002a. Suicide rates in China,1995–1999. Lancet 359, 835–840.

Phillips, M.R., Yang, G.H., Zhang, Y.P., et al., 2002b. Risk factors forsuicide in China: a national case-control psychological autopsystudy. Lancet 360, 1728–1736.

Robins, L.N., Regier, D.A. (Eds.), 1991. Psychiatric Disorders inAmerica: The Epidemiological Catchment Area Study. Free Press,New York.

Shi, Q.C., Zhang, J.M., Xu, F.Z., et al., 2005. Epidemiological surveyof mental illnesses in people aged 15 and older in ZhejiangProvince. Chinese Journal of Preventive Medicine 39, 229–236.

The WHO World Mental Health Survey Consortium, 2004. Preva-lence, severity and unmet need for treatment of mental disorders inthe World Health Organization world mental health surveys.Journal of the American Medical Association 291, 2581–2590.

Yang, G.H., Phillips, M.R., Zhou, M.G., et al., 2005. Understandingthe unique characteristics of suicide in China: National psycho-logical autopsy study. Biomedical and Environmental Sciences 18,379–389.

Yeh, E.K., Hwu, H.G., Chang, L.W., et al., 1985. Lifetime prevalence ofmental disorders in a Chinesemetropolis and 2 townships. In: Yeh, E.K., Rin, H., Yeh, C.C., et al. (Eds.), Prevalence of Mental Disorders.Department of Health, ROC, Taipei, Taiwan, pp. 175–197.

Zar, H.G., 1999. Biostatistical Analysis (4th edition), pp. 223–225,563–565. New Jersy: Prentice Hall.



Documents

Assessing depressive symptoms in persons who die of suicide in mainland China