All across the country, hospitals and doctors have been drawing new alliances to create sprawling “accountable care organizations” meant to make it easier for primary care doctors and specialists to work together to keep people healthy and avoid expensive hospitalizations.

But a study by a group of Boston researchers, published Monday in the Journal of the American Medical Association, found that, among some of the most expensive patients in the country, only a small percentage of hospital costs may be truly avoidable.

The researchers looked at the 10 percent of Medicare patients who account for 70 percent of the $91.7 billion in acute care costs for that program in 2010. Among that group, only 10 percent of hospital costs were the result of “preventable” conditions, such as hypertension, urinary tract infection, or uncontrolled diabetes.

Accountable care organizations should help in addressing preventable hospitalizations, by improving care coordination. But lowering the cost of “nonpreventable” conditions-stroke, sepsis, cancer, or dementia-requires a different approach, said Dr. Ashish Jha, lead author and professor of health policy and management at Harvard School of Public Health.

Reducing costs associated with hip replacement surgery, for example, might require finding cheaper, equivalent replacement devices or shortening the time a patient spends in the hospital, Jha said.

The authors note that bundling payments to doctors or hospitals-giving them one fee for all the care necessary to manage a particular condition-could push some of those changes.

“These findings suggest that strategies focused on enhanced outpatient management of chronic disease, while critically important, may not be focused on the biggest and most expensive problems plaguing Medicare’s high-cost patients,” they write.

When the paper was presented Monday at a research meeting, Jha said some in the audience were surprised that even 10 percent of hospital costs were truly preventable.

The same authors published another study Tuesday, in JAMA Internal Medicine, looking at whether performance indicators on three commonly-reported conditions-heart attack, heart failure, and pneumonia-give any indication about how well a hospital may do with unrelated medical conditions or procedures, such as stroke, sepsis, lung surgery, or amputation.

The group found that mortality rates for the reported conditions closely predicted mortality rates for other medical conditions and were somewhat less predictive for surgical procedures.

While an individual surgeon may affect the quality of a procedure, the authors wrote, “the broader findings from our work support the notion that there may actually be ‘good’ and ‘bad’ hospitals and that performance on a manageable set of key indicators can help identify such institutions.” Visit the Boston Globe for the article.