You are three weeks into your psychiatric rotation, and your preceptor turns a laptop toward you. On the screen is a free online calculator. You type in a handful of facts about a 34-year-old man with bipolar I disorder, and it spits out a single number: 0.8 percent chance of dying by suicide in the next twelve months. Your preceptor asks, “So, do we discharge him?”
That moment is coming for you, and what you do with that number is going to say a lot about the clinician you become. In its October 2026 issue, The Carlat Psychiatry Report ran a Research Update by Kamron Fariba, MD, on exactly this kind of tool: OxMIS, a free online suicide assessment calculator for patients with bipolar disorder or schizophrenia-spectrum illness. The article evaluates how well the calculator actually predicts. That is the right question, and it is one you should be able to answer before you ever click “calculate” on a real patient.
What OxMIS Actually Is
OxMIS stands for the Oxford Mental Illness and Suicide tool. It was built by Seena Fazel’s group at Oxford using a large Swedish national cohort, and it estimates the probability of death by suicide within twelve months for people aged 15 to 65 who carry a diagnosis of schizophrenia-spectrum disorder or bipolar disorder. According to the tool’s own user guide, it runs on 17 predictors covering sociodemographic, family, clinical, and violence-history factors, most of which are routinely documented in a chart. If an item is unknown, the calculator returns a range instead of a single number.
The user guide is also refreshingly candid about its limits. It says the tool performs most strongly at identifying people in the 0 to 1 percent risk range, which means it is best suited to ruling out the lowest-risk patients. It describes itself as an adjunct to full clinical assessment, not a replacement for clinical judgment, and it is meant to be used by healthcare staff because scoring some items takes interpretation.
Read that again, because that is the whole lesson in two sentences. A tool built to tell you who is at the low end of the risk curve is not the same thing as a tool that tells you who is about to die.
The Numbers You Need to Know Cold
The most useful independent test of OxMIS so far came from Finland. In a 2023 paper in Translational Psychiatry, Sariaslan and colleagues, including Fazel and Fanshawe, validated the tool in 137,112 people with severe mental illness drawn from national registers. Within twelve months of assessment, 1.1 percent of them, 1,475 people, died by suicide. The tool’s discrimination was an AUC of 0.70, with a confidence interval of 0.69 to 0.71. At a 1 percent predicted-risk cutoff, sensitivity was 58.8 percent and specificity was 70.3 percent. The authors also found the model overestimated risk for people with predictions above 5 percent, which is why they recommended reporting those results simply as “greater than 5 percent.”
An AUC of 0.70 is respectable for a prediction tool in psychiatry. It is also nowhere near what most people imagine when they hear the word “calculator.” So let’s do the math your boards will want you to do.
Imagine 10,000 patients with bipolar disorder or schizophrenia, with the same 1.1 percent twelve-month suicide rate as the Finnish cohort. About 110 will die. Using the published sensitivity, the tool flags roughly 65 of those 110 as high risk and misses about 45. That is a miss rate of around four in ten. Now look at the other side. Of the roughly 9,890 who will not die, a specificity of 70.3 percent means about 2,940 get flagged anyway. So of the roughly 3,000 patients flagged at that cutoff, only about 65 will die. That is a positive predictive value of about 2 percent, which is my own arithmetic from the published figures, not a number the authors reported. The vast majority of people the tool flags will not die by suicide within the year, and a meaningful share of people who do die will not have been flagged at all.
If that sounds like it contradicts the tool being useful, it does not. It means the tool answers a narrow question well, and a different question poorly.
Why Sensitivity, Specificity, and Base Rates Are Not Just Exam Trivia
Every PMHNP program teaches screening versus diagnostic testing, and every board prep book has a question or two where you have to reason about predictive value. Suicide risk is the clinical situation where those concepts stop being abstract. Suicide is rare, even in high-risk populations. When the base rate is low, even a decent test generates a pile of false positives. That is not a flaw in OxMIS. It is arithmetic.
This is also why the user guide’s emphasis on the 0 to 1 percent range matters. A tool that is good at separating “very low” from “not very low” can help you decide where to spend scarce time and scarce beds. It cannot tell you that this particular patient, sitting in front of you today, is the one in a hundred.
There is a second concept worth locking in: calibration versus discrimination. Discrimination is how well a tool ranks people from lower to higher risk, which is what the AUC measures. Calibration is whether a predicted 3 percent really means 3 out of 100. The Finnish team found OxMIS was well calibrated when predictions were capped at 5 percent, but it overestimated risk above that threshold. A tool can rank patients sensibly and still hand you a misleading number. Boards love this distinction, and so do patients’ families when they ask, “What are the odds?”
Screening Is Not Prediction, and the Joint Commission Says So
Here is where the clinical practice side comes in. The Joint Commission’s 2023 Quick Safety advisory on validated suicide risk screening tools makes a distinction every new prescriber should tattoo on the inside of their eyelids. Screening identifies people who may need further evaluation. It is not a complete determination of risk. The advisory says a positive screen should be followed by an evidence-based assessment process, used together with clinical evaluation, and it stresses that clinical staff who screen should be trained and assessed for competence in using the tool and in knowing what to do with the result.
The Joint Commission requires hospitals to screen patients being treated for a behavioral health condition as their primary reason for care, and behavioral health organizations must screen everyone they serve for suicidal ideation. Those screening tools, such as the Columbia Suicide Severity Rating Scale, answer a different question than OxMIS does. They ask whether this person is thinking about suicide right now. OxMIS asks, given who this person is on paper, how does their long-run twelve-month risk compare to people with similar histories. One is a snapshot of the present. The other is a statistical portrait of the year ahead.
What the Feasibility Work Tells Us
Before the Finnish validation, Senior and colleagues tested whether OxMIS could be run on real-world health records at all, publishing their results in Frontiers in Psychiatry in 2020. They manually scored 57 patients from electronic health records and found estimated twelve-month risks ranging from 0.1 percent to 3.4 percent, with a median between about 0.5 and 0.7 percent. Thirteen of the 17 predictors were complete in every record. Four were not: education, parental drug or alcohol use disorder, receipt of benefits, and parental psychiatric admission. Those are exactly the kinds of facts that never make it into a rushed intake note.
That finding is quietly one of the most practical in the whole story. Your calculator is only as good as your history. If you skip the family history and the social history, the tool either gives you a range so wide it is useless or, worse, you click “no” on everything you did not ask about.
A 2025 mini-review and meta-analysis in the Journal of Affective Disorders has since pulled together the validation evidence on OxMIS. I could not read its full results for this post, so check the abstract yourself, but the field-level takeaway echoes what the tool’s own authors say: encouraging predictive value and feasibility, with a continuing need for external validation in more countries and more care settings.
How This Shows Up on Your Boards
You will not be asked whether OxMIS exists. You will be asked questions built on the same logic, and a few patterns are worth practicing.
First, expect a vignette in which a patient scores “low risk” on a tool and also has a recent loss, new akathisia, command hallucinations, or a firearm at home. The correct answer will not be “discharge, the score was low.” Tools inform, risk factors and warning signs still drive the plan.
Second, expect a stem about a screening test with a given sensitivity and specificity in a low-prevalence population, asking what is most likely true about its positive predictive value. Now you know the answer is “low, because the condition is rare,” and you know the clinical stakes behind the question.
Third, expect questions that test whether you know the difference between a risk factor, a warning sign, and a protective factor, and whether you can build a safety plan that actually reduces access to means. The calculator does none of that work.
Using a Tool Like This Without Letting It Use You
So what would a thoughtful new PMHNP do with a free calculator in a busy clinic? Here is a way of thinking about it that holds up under pressure.
Use it for the population it was built for. OxMIS is validated for ages 15 to 65 with schizophrenia-spectrum or bipolar disorder, and it was derived in Sweden and tested in Finland. Do not stretch it to a patient with major depression, borderline personality disorder, or a first-episode substance-induced psychosis, and be humble about how well a Scandinavian registry population maps onto your patients.
Use it to organize your thinking, not to end it. A low number can help you justify spending your most intensive resources elsewhere, which is a real clinical need. A higher number is a reason to look harder, never a reason to feel certain.
And document your reasoning. The standard of care is not “I ran a calculator.” It is “I assessed, I considered the tool’s result in context, I made a plan, and here is why.”
The Bigger Picture for Your Training
Dr. Fariba’s piece in the October Carlat issue sits inside a larger shift. Psychiatry wants better prediction, and rightly so, because the field’s track record at forecasting suicide from clinical impression alone is humbling. Free, transparent, peer-reviewed tools like OxMIS are a step toward something better. They are also a test of whether clinicians can hold two ideas at once: the tool is useful, and the tool is not enough.
That ability, to use evidence without being hypnotized by it, is the single most valuable habit you can build in graduate school. It will serve you on exam day, when a question gives you a sensitivity and a prevalence and asks you to think. It will serve you in a crowded clinic, when a number on a screen tempts you to skip the conversation. And it will serve you on the day a patient looks at you and says, “I’m fine,” when everything else in the room says otherwise.
You are not studying to become a human calculator. You are studying to become the person who knows what the calculator cannot see.
Ready to turn concepts like sensitivity, predictive value, and risk assessment into points on exam day? Head over to PMHNPReview.org for board-focused study resources built for students exactly like you. You have the compassion for this work. Let’s make sure you have the preparation to match it.