Although the healthcare community has expressed a strong desire to measure safety outcomes, accomplishing this feat has been challenging due to poor investment in the basic science of patient safety. There is a need for basic science because it can allow for better understanding of the causes of harm, help in designing and pilot testing interventions to reduce harm, and enable researchers to evaluate the effects of harm. To advance the science of measuring safety outcomes, it’s critical to separate and distinguish preventable harm from inevitable harm.

In healthcare, the term preventable harm differs substantially from that in other industries. Despite receiving evidence-based medical therapies, some patients will inevitably die or sustain complications and problems that are preventable are likely to change over time. It’s important to consider strategies that tease apart preventable harm from inevitable harm, such as:

Assuming all harm is preventable (high sensitivity, low specificity).
Adjusting for preventability (low sensitivity, low specificity).
Linking care received to outcomes (high specificity, low sensitivity).

Assessing Potential Strategies

Virtually all harm has been labeled as inevitable for decades by clinicians, but recent efforts by payers (eg, CMS) have aimed to label all harm as preventable. This strategy could be appropriate when evidence suggests that most harmful events are preventable. However, the problem is most measures of harm are missing one or several of the required validity components. Most harms are preventable to some degree, but we don’t have evidence to tell us how much.

Another strategy could be to use risk-adjustment models to account for preventable and inevitable harm. Such models typically adjust for severity of illnesses, patient demographics, comorbid conditions, and diagnoses, but they do little to motivate efforts to improve care. Surveillance bias and measurement and random error can influence accuracy. Furthermore, these models often support current performance and maintain the status quo.

Linking the care received to adverse outcome measures is another approach. For example, outcomes would be labeled a preventable harm if evidence-based therapy or standards were not rendered or were rendered incorrectly and the patient sustained an adverse outcome. Ideally, healthcare organizations could then monitor event rates of these events and payers could create financial incentives to minimize these events. Unfortunately, a precondition of this process-outcome model is that evidence or standards exist on therapies that can prevent the harm, but this isn’t always the case. For example, this method would fail to capture some harmful events that resulted from poor teamwork or other communication errors.

Methods to Move Forward

Given the risks and benefits of these different strategies, science should guide policy. The federal government should invest in the basic science of patient safety to develop scalable measures of preventable harm. Measures should be meaningful to clinicians who will use them to improve care and then aggregate measures to the health system, state, and national levels. It’s also important to make estimates of measurement error transparent while recognizing that tradeoffs must be made between accuracy and costs. Separating hospital efforts that prevent harm from policy efforts that judge performance will also be helpful, and new interventions should be identified to prevent harm. Efforts undertaken without the evidence that’s needed may do more harm than good. Once healthcare can accurately estimate the extent to which harm can be prevented, policy makers can then align payment policies accordingly.