Predicting intimate partner violence before a patient speaks to a doctor has shifted from academic debate into clinical modeling in major research hospitals. In a recent preprint study that has not yet undergone peer review, investigators Eden Hoffer and Karen Campbell evaluated these new predictive systems [1]. Working with teams led by Harvard Medical School in Boston, researchers trained artificial intelligence algorithms on routinely collected medical files to identify subtle risk patterns during routine clinic visits [2]. In hospital evaluations, the software flagged people at elevated risk more than three years before they enrolled in a domestic abuse intervention program. Boston clinicians tracked patient outcomes across multiple cohorts [1].
Predictive Models for Intimate Partner Violence
The Harvard Medical School team built their predictive tool by analyzing routinely collected electronic health records from female patients who sought care at a single American academic health centre located in Boston. Boston records spanned several years of patient history [1]. In an analysis published through The Conversation, Eden Hoffer and Karen Campbell explained that algorithms scanned diagnostic codes, clinical notes, and prescription histories to identify subtle combinations of health complaints that preceded formal disclosures of intimate partner violence in three separate hospital validation cohorts that researchers tested in Boston health networks [2]. Yet the Harvard Medical School authors added a clear note of caution, warning that “careful performance evaluation on the more general population of female patients should be conducted before implementing the models for clinical use” [1].
Predictive screening tools often promise early detection, but they can inherit systemic distortions from older clinical files. These patient records often underrepresent vulnerable groups or reflect unequal diagnostic practices, raising valid concerns about algorithmic bias in clinical data and automated hospital alerts when algorithms guide bedside decisions. Missing diagnostic records hide lived pain. Developers at Harvard Medical School said the software aims to help clinicians talk with kindness rather than pushing people to disclose private trauma [2].
Computer predictions cannot ensure personal safety once a patient walks out of the examination room. Community shelters face long service backlogs and funding shortages that computer code cannot fix [2]. Without adequate staff training, early alerts risk leading to intrusive questions that make patients feel cornered during routine medical visits [1].
Algorithmic Flags and Survivor Agency
Viewing intimate partner violence through the lens of trauma- and violence-informed care reveals deep ethical tensions that technical benchmarks overlook [1]. As Eden Hoffer and Karen Campbell outlined in The Conversation, this clinical framework emphasizes core principles of safety, choice, collaboration, connection, and capacity-building to support healing rather than imposing outside control on people who live in danger [2]. Deciding whether to disclose relationship abuse involves complex personal assessments of physical safety, housing stability, and financial dependence [1]. For many women, remaining silent about intimate partner abuse is a survival strategy to protect themselves and their children [2].
What happens when a health system claims to know about intimate violence before a patient chooses to speak? An automated alert created by predictive algorithms can strip away personal autonomy, taking away a survivor’s freedom to decide when, where, and how to discuss traumatic relationship experiences with a trusted doctor [1]. Abuse centers on control and coercion [2]. When software infers private trauma from electronic health records, it risks overriding the right of survivors to direct their own care [1].
Automation bias presents another serious hazard in modern medicine. Research on clinical decision-support systems shows that doctors can put too much faith in computer predictions, missing critical emotional cues when an automated model fails to generate a risk warning during an appointment [1]. As Eden Hoffer and Karen Campbell warned, the absence of an alert might discourage clinicians from asking compassionate questions directly [2]. Machine tools cannot replace human clinical empathy [1].
Mental Health Burdens of Domestic Violence
The lasting consequences of intimate partner abuse extend far beyond bodily injuries, shaping psychological health across adult lifespans [3]. In a study published in the Journal of Family Violence, a research team led by Megan Willis analyzed responses from over 7,000 Australians who had experienced an intimate adult relationship [4]. Drawing on data from the nationally representative Australian Child Maltreatment Study, the researchers evaluated four specific conditions: major depression, generalized anxiety disorder, alcohol abuse, and post-traumatic stress disorder (PTSD) [3]. Megan Willis documented high rates of psychiatric illness among survivors [4].
Survey results showed that 55.3% of relationship violence survivors met diagnostic criteria for at least one measured psychological condition [3]. By comparison, only 24.5% of participants who had never experienced partner violence met those same clinical criteria. The measured difference between the two groups was more than thirty percentage points. As Megan Willis reported in the Journal of Family Violence, rates of mental illness remained high across all three measured categories of harm, showing that physical assaults, sexual violence, and psychological abuse often occur together and reinforce long-term emotional distress for victim-survivors over time [4].
Childhood adversity compounds these adult psychological harms [3]. As reported in the Australian Child Maltreatment Study, 62.5% of individuals who grew up with multiple forms of child abuse qualified for a psychiatric diagnosis, compared with 39.1% among survivors who reported no childhood trauma in their survey responses [4]. This cumulative vulnerability shows why healthcare providers must review a patient’s complete life history during clinical assessments of violence risk rather than relying solely on isolated electronic health codes [3].
Why Subpoenaed Records Threaten Clinical Privacy
Many abuse survivors struggle to get professional help [4]. Findings from the Australian Child Maltreatment Study revealed that approximately two in five survivors exposed to physical or sexual aggression never seek medical treatment or counseling advice [3]. Local domestic violence services struggle to meet surging community demand, while out-of-pocket medical expenses keep broader psychological therapy out of reach for many families who need ongoing emotional support [4]. Steep therapy fees often stop people from seeking help [3].
Legal hazards create another formidable barrier that keeps survivors from speaking candidly with health care practitioners [4]. In contested custody disputes examined by researchers in Medical Xpress, confidential medical notes, psychiatric evaluations, and therapy records can be subpoenaed and inspected by legal teams representing an abusive former partner who seeks to undermine the parent’s credibility [3]. Custody data reveal that domestic abuse claims arise in more than 80% of parenting matters heard by Family Court judges. Practitioners have described these broad legal demands as “fishing expeditions” that can be weaponized as a form of systems abuse in Family Court cases. Megan Willis emphasized that seeking mental health support should never compromise parental rights in Family Court proceedings [4].
Although Australian family law reforms introduced in 2025 strengthened statutory protections for confidential therapy records, those legal safeguards do not apply automatically across court proceedings [3]. Statutory protections need accessible legal aid. Without lawyers to invoke privacy protections, many parents see their private therapy notes disclosed in court, discouraging patients from seeking mental health treatment after surviving domestic abuse [4]. Legal exposure discourages honest talks with doctors [3].
Early Psychological Signs of Partner Violence
Abusive dynamics usually begin with quiet manipulation [4]. In data from the Australian Child Maltreatment Study, roughly a third of women lived with emotional abuse while reporting no physical or sexual violence from an intimate partner [3]. When individuals experienced multiple forms of mistreatment over time, psychological control almost always served as the initial phase [4]. Isolating a partner from family, monitoring daily movements, and employing verbal degradation establish an ongoing pattern of coercive control [3].
Because psychological abuse often masquerades as protective affection, victim-survivors frequently struggle to recognize controlling behavior during the opening stages of an intimate relationship [4]. In reports published in Medical Xpress, Megan Willis explained that jealous outbursts, possessiveness, and demands to restrict personal friendships are sometimes misinterpreted as passionate devotion rather than early warning signs of escalating coercive control that threaten long-term safety [3]. Controlling conduct rarely fades away on its own. Public health awareness efforts need to describe clearly what psychological abuse looks like in everyday life so people seek support sooner [4].
Clinical identification tools must detect subtlety [1]. If hospital algorithms only register obvious physical injury codes, they miss the formative years when psychological control dominates the relationship and inflicts lasting psychiatric harm on vulnerable patients [2]. As researchers from Harvard Medical School explained, detecting subtle indicators early provides vital opportunities to offer safety planning, counseling resources, and community domestic abuse referrals before acute physical crises occur in the household [4]. Timely clinic advice changes lives [3].
Detecting IPV and Meeting Survivor Needs
Automated detection does not ensure genuine recovery. In clinical perspectives from Harvard Medical School, researchers noted that the development of AI-generated risk models reveals an essential distinction: detecting IPV through automated software records is fundamentally different from helping an individual heal from relationship trauma [1]. Even if an algorithm correctly identifies elevated risk, health systems often lack the emergency housing, legal aid, and long-term counseling services necessary to keep a survivor safe once they leave the hospital. Community resource deficits weaken predictive alerts [2].
Connecting a patient with effective support requires trained healthcare workers who understand how to respond to disclosures of abuse with compassion, respect, and trauma-informed clinical care [4]. Human clinical judgment guides sensitive doctor visits. Without sufficient community resources and trained personnel, automated alerts merely document vulnerability without offering practical pathways to physical safety or long-term personal stability for victim-survivors [2]. Writing in The Conversation, clinicians Eden Hoffer and Karen Campbell said that AI-based IPV tools should supplement clinical judgment rather than dictating medical decisions [1].
Interactions within health care settings provide an important window for identifying intimate partner abuse and extending compassionate assistance to those in danger [1]. Technology must serve survivor-centered agency [2]. Predictive tools should foster trusting conversations between patients and doctors rather than deciding when a survivor’s private story is exposed to the wider healthcare system [1]. Real healing begins when healthcare providers listen to survivors on their own terms, protecting personal agency throughout the recovery process [2].
- PREPRINT Hoffer, E., & Campbell, K. (2026). AI can predict intimate partner violence before survivors disclose. But should it? [Article Link]
- ONLINE NEWS Hoffer, E., & Campbell, K. (2026, October 14). AI can predict intimate partner violence before survivors disclose. But should it? The Conversation. [Article Link]
- ACADEMIC JOURNAL Beaton, T., Higgins, D. J., Mathews, B., Hegarty, K. L., Scott, J. G., & Willis, M. L. (2026). Prevalence of Mental Disorders Associated with Intimate Partner Violence Victimisation by Gender, Sexuality, and Child Maltreatment History. Journal of Family Violence. [Article Link]
- ONLINE NEWS Willis, M. (2026, October 14). Mental disorders are twice as common among victim-survivors of intimate partner violence. Medical Xpress. [Article Link]
APA 7: TWs Editor. (2026, October 15). AI Models Predict Intimate Partner Violence Years in Advance. PerEXP Teamworks. https://perexpteamworks.com/en/intimate-partner-violence-ai-prediction/
