1. Screening for Atypical Suicide Risk Using Rasch Person Fit Statistics Ken Conrad, University of Illinois at Chicago Nikolaus Bezruczko, Independent Consultant HyeJung Park, Barth Riley, University of Illinois at Chicago Ya-Fen Chan, Michael Dennis, Chestnut Health Systems, Bloomington, IL

2. Can fit statistics discern a clinical profile? Using the modern psychometric technique of person fit statistics, this study attempted to discern a profile of atypical suicide risk. Such high-risk but atypical persons were defined as those who scored high on symptoms of suicide risk but did not fit the expectations of the Rasch measurement model (Rasch, 1960; Wright & Stone, 1979).

3. Aha! Eureka! Gold in them hills? The topic of this paper arose when, during a Rasch analysis of a measure of psychological distress*, the fit statistics indicated that misfitting persons were endorsing high severity suicide items but not the less severe items such as depression. This is contrary to the expectation of data fitting the Rasch Model. *Internal Mental Distress Scale of the Global Individual Severity Scale (GISS) of the Global Appraisal of Individual Needs (GAIN, Dennis, et al., 2003). Please see handout.

4. Current Theory of Suicide The existing literature indicated a fairly strong consensus that there were two general types of persons at high risk for suicide. The more typical type had depression and intense psychological pain that persisted over time. The second more atypical type was believed to be impulsive and somewhat prone to violent behavior.

5. SEVERITY PERSONS* DEPRESSIONSOMATICANXIETYTRAUMASUICIDE 2.5. | Means. |T ShudBpunish. | Attempts 2. T+ Plans. | AfradSleep. |. | OthsNoSee. |S WishDead 1.# + SnakesDark WakColdSweat. | FearOpenSPac CantGoOn.# | MuslAches NoFeelingSuicideThgts.## | MovSlowerDryMouth HrtSomeone.# S|.# | Worries Nitemares 0.## +M LostCool.# | AnyAboveProb.# | HeadachFaint Trembling.## | PainChest AODHlpSleep.### | TakAdvantage.# | Shyness RemindDistrs FeltGuilt -1.### + LostInterest.## |S Anxious.## M| DontUnderstnd.## | Sadepresssed SleepTrouble.# |.### | -2.### +.### |. |T AnoyedIritabl -2.5.#### | * Each # is 43 people and each. is 1-42 persons. Chart was truncated and does not show 6 people higher than all items or 1144 people below all items. IMDS person reliability is.89. Remembering Tired Wghtloss Repeatover Arguments OthrsWatch NoXpressFlngs FearUrges Person Item Map of IMDS & Subscales

6. The Concept of Fit in Rasch Model: The Expectation The Rasch model expects endorsement of items of increasing severity to be associated with increased pathology. e.g. psychological distress, where a 1 refers to endorsement, agreement, or being correct whereas 0 refers to lack of endorsement, disagreement, or being incorrect. Less 1 1 1 1 1 1 1 0 1 0 1 0 1 0 0 0 0 0 0 0 More The line of 1s above indicates items that are easy for the person to endorse, while the line of 0s indicates items that are harder to endorse, i.e., do not apply to the person. The zone of uncertainty, where 0s and 1s are equally probable, indicates the persons level or amount.

7. Example of Good Fit: 1.0 mnsq is expected For example, in the assessment of depression, an item such as I wish I were dead, represented by the enlarged and bolded 1 below, is typically a high severity item that may be endorsed by people who have already endorsed less severe symptoms such as feeling depressed, loss of interest, and being afraid to go to sleep. Less 1 1 1 1 1 1 1 1 1 1 0 1 0 1 0 1 0 More

8. Poor Outfit mnsq (>1.33): where a person did not endorse the less severe such as depression & loss of interest (zeroes at left), but did endorse high severity symptoms I wish I were dead, the bolded 1, and other suicide items, the other 1s. Less0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 More Therefore, the pattern above is interpreted as a person at high risk for suicide, but who does not fit the typical profile expected both by theory and by the Rasch model. The Rasch model would flag this pattern as highly unexpected or misfitting. A person with this pattern would score low, like someone with mild depression, on the overall measure even though they endorsed several of the most severe items. Therefore, the ability to flag such persons would be useful clinically.

9. Over-fitting, <.75 mnsq The most typical or prototypical persons would fit the model too well or would be unexpectedly perfect: Less1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 More The above pattern is judged methodologically as fitting too well because the model expects some randomness or uncertainty around the persons level on the trait, whereas here the level is pinpointed exactly.

10. Poor Infit: Illogical Inlying Pattern Less 0101010101010101010101 More Infit mnsq > 1.33

11. Suicide Items MOST MISFITTING RESPONSE STRINGS 16. Thoughts of suicide OUTMNSQ |ITEM 17. Plan to commit |1 1 222 11 3424 3 2443231 1331332323241121 18. Got means to carry out |02157901362382362491240173534506874654817998 19. Attempted suicide high-------------------------------------------- 9.90 A|....................................1. Scale: Yes=1, No=0 9.58 B|..............................1........11.. on all IMDS items 8.72 C|......................... 1.........1.. 8.06 D|....................................1. 7.98 E|......................1.. 1........1..1 5.96 F|..............................1.1......11.1 5.72 G|......... 1............... 1.........1.. 5.35 H|......................... 1........11.1 5.32 I|...............................1...... 4.91 J|...................................1....... 4.91 K|...................................1....... 4.91 L|...................................1....... 4.90 M|............1.........................1.... 4.76 N|..............................1........11.1 4.65 O|...0.0......................0......... 4.24 P|.......................1......1........11.1 4.14 Q|......................1.. 1........1..1 3.90 R|..............................1......1.11.1 3.85 S|.................................1......... 3.85 T|.................................1......... 3.70 U|..........1.......1..1... 1.........1.. 3.66 V|...................1.....1...............1. 3.61 W|.........................1.........1....... 3.60 X|..............................1........11.1 3.59 Y|.......................1. 1.....1...1.. 3.55 Z|..............1...1....1...........1.....1. |----------------------------------------low-

12. VariablesMost typical <.75 outfit MNSQ Typical.75 – 1.33 Atypical >1.33 outfit MNSQ SuicideSimilar, high DepressionHighestMediumLowest SomaticHighestMediumLowest AnxietyHighestMediumLowest TraumaHighestMediumLowest Substance ProblemsSimilar, high Internal Disorders, e.g., inattentive, impulsive LowestMediumHighest External Disorders, e.g., Crime & Violence LowestMediumHighest Outfit Groups, i.e., how atypical patterns of responses will compare to typical patterns on variables of interest

13. Comparison of Suicide Group vs. No Suicide Group -2.5000 -2.0000 -1.5000 -0.5000 0.0000 0.5000 1.0000 1.5000 2.0000 2.5000 IMDSSuicideDepressionAnxietyTraumaSomaticSPSBCSCrime&Violence Other GISS Scales Logits No Suicide (N=6,165) Suicide (N=1,193) IMDSIMDS Subscales

15. Outfit Group Mean/Total % VariableLow N=176 Medium N=546 High N=471 Mean Age (SD)24.59 (11.47)20.49 (8.92)19.90 (9.40)20.86 (9.65) Employed*42.9%40.9%33.4%38.2% Diagnosed w. Mental Disorder100.0%98.2%94.3%96.9% Cur. Crim. Just. Involvement55.7%60.3%62.0%60.3% Fam. Hist. Psychological Prob.70.2%62.6%60.6%63.0% Fam. Hist. Drug Use70.2%72.0%74.3%72.6% Past Yr Suicide Any Symptoms100.0% Suicidal Thoughts100.0% 99.8%99.9% Got gun to carry out plan**17.0%21.1%53.5%33.3% Had a plan to commit suicide**23.9%28.0%68.8%43.5% Attempted Suicide**19.9%25.5%61.8%39.0% Past Yr Depression Feeling Tired/No Energy a** 100.0%79.0%55.0%71.0% Moving/Talking Slower b** 59.2%33.7%27.0%33.6% No Energy/Losing Interest in Friends/Work c** 98.1%72.6%61.4%73.6% Note: a Based on a total number of cases = 662 b Total number of cases = 561 c Total number of cases = 531 Bivariate analyses with outfit groups ª in the subset with suicidal ideation

16. Conclusion With the computer-administered GAIN, it was possible to quickly identify persons with suicidal ideation. Using Rasch analysis of person fit statistics, it was possible to clarify rapidly which were most typical, typical and atypical. We believe that these capabilities are useful advances over unstructured interviews or paper and pencil questionnaires that are commonly in use. Using a large sample, we confirmed the expectation that the atypical profile was associated with relatively low depression. However, unexpectedly, the atypical profile was not strongly associated with internal disorders such as inattentiveness and impulsivity or external disorders such as committing crimes and being violent.

17. Future Directions Perhaps most significant clinically is that this technique may be able to spotlight many persons at high risk for suicide who are least likely to be identified using the usual screening methods. Future research should further explore the usefulness of Rasch person fit statistics in research on suicide as well as in other clinical applications.