New research from a major 'randomised' arrest experiment 23 years ago finds that domestic violence victims whose partners were arrested on misdemeanor charges - mostly without causing injury - were 64% more likely to have died early, compared to victims whose partners were warned but not removed by police.

Among African-American victims, arrest increased early mortality by a staggering 98% - as opposed to white victims, whose mortality was increased from arrest by just 9%. The research also found that employed victims suffered the worst effects of their partners' arrests. Employed black victims with arrested partners suffered a death rate over four times higher than those whose partner received a warning at the scene. No such link was found in white victims.

The study's authors say that causes are currently unknown but such health impacts are consistent with chronic stress that could have been amplified by partner arrest. They call for a "robust review" of US and UK mandatory arrest policies in domestic violence cases.

"It remains to be seen whether democracies can accept these facts as they are, rather than as we might wish them to be," said Professor Lawrence Sherman from Cambridge University's Institute of Criminology, who authored the study with his colleague Heather M. Harris from University of Maryland.

The findings were announced in the US on Monday 3rd March in Milwaukee and College Park, Maryland, and presented on Wednesday in London at the winter meeting of the Society of Evidence-Based Policing. Milwaukee Police Chief Edward Flynn, who supported the follow-up study, will join in the presentation and discussion of the results. The study will be published in a forthcoming edition of the Journal of Experimental Criminology.

The vast majority of victim deaths following the Milwaukee Domestic Violence Experiment were not murders, accidents or suicides. The victims died from common causes of death such as heart disease, cancer and other internal illnesses.

Previous studies have shown post-traumatic stress symptoms (PTSS) to be prevalent in victims of domestic violence, and that low but chronic PTSS has been linked to premature death from coronary heart disease and other health problems. The authors observed that the impact of seeing a partner arrested could create a traumatic event for the victim, one that raises their risk of death. An arrest may cause more trauma in concentrated black poverty areas than in white working-class neighbourhoods, for reasons not yet understood.

The exact cause of these surprising results still remains a 'medical mystery,' the study's authors say. But, whatever the explanation, the harmful effects of mandatory arrest poses a challenge to policies that have "been on the books" in most US states and across the UK for decades, they say.

"The evidence shows that black women are dying at a much higher rate than white women from a policy that was intended to protect all victims of domestic violence, regardless of race," said Sherman. "It is now clear that a pro-arrest policy has failed to protect victims, and that a robust review of these policies is urgently needed."

"Because all the victims had an equal chance of having their partners arrested by random assignment, there is no other likely explanation for this difference except that it was caused by seeing their partners arrested."

The Milwaukee Domestic Violence Experiment took place between 1987 and 1988, with support from the National Institute of Justice, the research arm of the US Department of Justice. Sherman, who led the study from the University of Maryland, described it as "arguably the most rigorous test ever conducted of the effects of arrest."

The experiment enrolled 1,125 victims of domestic violence whose average age was 30 years. Each case was the subject of an equal probability 'lottery' of random assignment. Two-thirds of the suspects were arrested with immediate jailing. One-third received a warning at the scene with no arrest. In 2012-13, Sherman and Harris searched state and national records for the names of every one of the victims.

The record search showed that a total of 91 victims had died. Of these, 70 had been in the group whose partners were arrested, compared to 21 whose partners had been warned. This translated into 93 deaths per 1,000 victims in the arrest group, versus 57 deaths per 1000 in the warning group. For the 791 black victims (who were 70% of the sample), the rates were 98 per 1,000 for arrest, versus 50 per 1,000 for the warned group.

"These differences are too large to be due to chance," Sherman said. "They are also too large to be ignored."