Kosovo and Romania Show the Gap. Can a Questionnaire Tell Which Women Are About to Be Killed?
A woman in Kosovo was killed hours after reporting violence, and Romania's new femicide law still lacks the tool meant to spot women in danger. What lethality risk assessment is, what the research says it can and cannot predict, how Spain's VioGén algorithm, Britain's DASH form, Brazil and India do it, and why a score is only worth the protection it triggers. Explainer with a short opinion section.
On Sunday 25 October, Arlinda Zekaj, a 49-year-old woman from Vitomirica near Peja in Kosovo, went to the police to report domestic violence. Her ex-husband was arrested and held for 48 hours. Police escorted her home. Around 9 p.m. she was killed in that home, and the suspect is her brother-in-law, who is still on the run, according to CNA. Three days earlier she had asked the prosecutor to act because, as Balkan Insight reports, she feared for her life. She was the eighth woman killed in Kosovo this year, and in every case a spouse or family member is suspected.
The same week in Bucharest, a parliamentary committee heard that Romania’s femicide law has been in force since April but still lacks the methodology meant to help authorities spot women at high risk of being killed. Institutions appeared to pass the job between them, RO Times reported, and the Justice Ministry had only just signed an order to start collecting the data the tool needs. Digi24 counted 30 femicides in Romania by early October.
Both stories point at the same question: can the state tell, at the moment a woman asks for help, whether she is about to be killed?
The scale, and the promise
Worldwide, about 50,000 women and girls were killed by partners or relatives in 2024, roughly 137 a day and 60% of all intentional killings of women, according to UNODC and UN Women. Unlike most homicides, many of these are preceded by a history the authorities already know about. That is the promise of risk assessment: if the warning signs are visible, the most dangerous cases can be sorted from the rest and protected first.
In Europe this is a legal duty. Article 51 of the Istanbul Convention requires authorities to assess “the lethality risk, the seriousness of the situation and the risk of repeated violence”, and to check at every stage whether the perpetrator has access to firearms.
What the research says predicts murder
The best-known tool is the Danger Assessment, developed by the American nurse-researcher Jacquelyn Campbell. In a 12-city US study comparing women killed by partners with women who were abused but not killed, use or threatened use of a weapon was the strongest marker: those women were about 20 times more likely to be among the dead. Women whose partner had threatened to kill them were almost 15 times more likely. The abuser’s gun ownership, especially once the couple lived apart, and his heavy drinking or drug abuse also set the killers apart. A separate analysis found that a past non-fatal strangulation raised the odds of later homicide about 7.5 times.
When the revised questionnaire was tested on an independent group of women who had survived attempted murder, it scored about 0.90 on the standard accuracy measure (area under the curve, where 0.5 is a coin toss and 1.0 is perfect), the developers reported. They also found it performed better than the women’s own sense of danger.
What happens when police use it
Research results and police practice are not the same thing. In England and Wales, most forces use the DASH checklist of 27 questions. The largest European study of it found officers’ risk ratings were “little better than random”, with accuracy scores around 0.6. In Dorset, researchers found that 45 of 67 deadly or near-deadly cases with earlier police contact had not been rated high risk, while 99% of the cases that were rated high risk ended without serious harm. In Thames Valley, 55% of deadly cases had no recorded police contact at all, so no form could have flagged them.
Spain went furthest. Its VioGén system, used by police in most of the country since 2007, turns 35 questions into a rating from “negligible” to “extreme”, and the rating decides protection: an “extreme” case can mean round-the-clock police guard. Spain’s own validation work found accuracy of about 0.60 to 0.66 for repeat and serious violence, modest but better than chance. A 2024 New York Times investigation found that at least 247 women had been killed by partners after a VioGén assessment, and that in a judicial review of 98 of those killings, 55 women had been rated negligible or low risk. The same reporting credited the system with reducing repeat attacks overall. The Eticas Foundation’s external audit, done without access to official data, said officers kept the automatic score in 95% of cases and argued that the number of “extreme” ratings follows the resources available. The Interior Ministry’s head of research dismissed that audit because it did not use VioGén data, and a senior police officer told the BBC that the system fails sometimes but “the only guilty party is the person who killed”. Spain launched VioGén 2 in January 2025 with new indicators and recalibrated algorithms, and the BBC reported that the “negligible” category was to be abolished.
Beyond Europe and North America
Brazil made a national form law in 2021. Law 14.149 created the FONAR questionnaire, applied preferably by the civil police when a woman files a report, or by prosecutors or judges at first contact, and meant to guide protective orders, as the National Justice Council explains.
Most tools were built around an intimate partner. Yet the UN’s own figure counts partners and relatives together, and the Kosovo case is a reminder that the man in the cell is not always the only danger. In India, researchers adapted the Danger Assessment into two versions, one for partners and one for in-laws, because in many joint households the threat comes from the husband’s family too. A tool imported unchanged from one family structure may miss the danger in another.
The costs and the limits
Every threshold is a trade. Set it low and police are flooded with “high-risk” cases they cannot protect, as the 99% false-positive rate in Dorset shows; set it high and women like those in Spain’s judicial review fall below it. Police officers and ministers also face real constraints: detention limits such as Kosovo’s 48 hours, courts they do not control, and the presumption of innocence of the accused. Kosovo’s justice minister, Donika Gërvalla, said in parliament that Zekaj “has not addressed the Government”, a remark activists called victim-blaming; she later clarified that she meant ministers cannot overrule prosecutors and judges and that Zekaj was right to turn to them.
There is evidence that assessment combined with action helps. In an Oklahoma study of a programme in which police screen women at the scene with a short lethality questionnaire and link those at high risk to support services, women who received it reported less severe violence and more protective steps about seven months later than a comparison group. That is a study of one programme, not proof it works everywhere.
Opinion: a score is only worth the protection it triggers
I can’t know whether any questionnaire would have saved Arlinda Zekaj, and nobody should pretend otherwise. But the research is clear about some signs, such as threats to kill, strangulation and access to weapons, and a state that asks those questions in a structured way will miss fewer women than one that relies on an officer’s mood. Romania has the law but not the tool; Kosovo’s parliament has now called for a review of active high-risk cases within 72 hours. Both are steps.
The harder lesson from Spain and Britain is that the form is the cheap part. A rating means nothing if “high risk” leads to no guard, no shelter bed and no swift court order, and an algorithm can quietly turn a budget limit into a risk level. Risk tools should be validated on local data, audited by people outside the ministry, include relatives as well as partners, and treat a woman’s own fear as evidence rather than noise. Most of all, the people who score a woman as safe should have to explain why when she is not.
Sources
- CNA: Resolution approved to investigate the murder of Arlinda Zekaj (27 Oct 2026)
- Balkan Insight: Kosovo Woman Murdered Hours After Reporting Domestic Violence to Police (26 Oct 2026)
- Koha: Gërvalla clarifies her statement in the Assembly (26 Oct 2026)
- Prishtina Insight: Kosovo Civil Society Demands Institutional Accountability After Latest Femicide (Oct 2026)
- RO Times: Chamber of Deputies debates femicide law implementation (27 Oct 2026)
- Digi24: Legea femicidului, doar pe hârtie (3 Oct 2026)
- Romanian Ministry of Justice: Participation in the joint special commission 'Romania without domestic violence' (27 Oct 2026)
- UNODC: 137 women and girls killed every day by intimate partners or family members in 2024 (Nov 2025)
- Council of Europe: Istanbul Convention (CETS No. 210), Article 51
- Campbell et al.: The Danger Assessment - Validation of a Lethality Risk Assessment Instrument for Intimate Partner Femicide (2009)
- Campbell: Research Results From a National Study of Intimate Partner Homicide - The Danger Assessment Instrument (NIJ, 2004)
- Glass et al.: Non-fatal strangulation is an important risk factor for homicide of women (2008)
- Turner, Medina and Brown: Dashing Hopes? The Predictive Accuracy of Domestic Abuse Risk Assessment by Police (British Journal of Criminology, 2019)
- Chalkley and Strang: Predicting Domestic Homicides and Serious Violence in Dorset (Cambridge Journal of Evidence-Based Policing, 2017)
- Thornton: Police Attempts to Predict Domestic Murder and Serious Assaults - Is Early Warning Possible Yet? (2017)
- New York Times: An Algorithm Told Police She Was Safe. Then Her Husband Killed Her. (July 2024)
- BBC: Spain police algorithm said Lina was at 'medium' risk. Then she was killed (April 2025)
- Eticas Foundation: The External Audit of the VioGén System (archived)
- López-Ossorio et al.: Validation and Calibration of the Spanish Police IPV Risk Assessment System VioGén (2019)
- La Moncloa: Interior designs a new model of police response to gender violence, VioGén 2 (15 Jan 2025)
- Brazil: Lei nº 14.149/2021, Formulário Nacional de Avaliação de Risco
- CNJ: Formulário Nacional de Avaliação de Risco (FONAR)
- BMC Public Health: Adapted danger assessments for women in abusive relationships in India, partners and in-laws (2024)
- Messing et al.: The Oklahoma Lethality Assessment Study (Social Service Review, 2015)
Comments