The Most Famous Malpractice Cases Were Decided by Memory. Yours Doesn't Have To Be.

For most of the history of American medical malpractice litigation, the central question in a case was answered from memory. What happened at three in the morning? Who was called, and when? What did the covering physician actually know at the moment the order was written? Juries answered those questions by deciding which witness recalled the night more convincingly, often years after the night was over.

That is no longer the only way to answer them. The electronic record keeps its own account of the same events, written as they happened, by a system with no stake in the outcome. The audit trail records who opened the chart, what was entered, when the entry was saved, and what was changed afterward. In most modern cases that information already exists. It is simply not always requested, and it is very often resisted.

Three of the most widely discussed cases in American patient safety make the point. Two were argued largely over reconstructed human recollection, because in their era that was the only evidence there was. The third was assembled from machine-generated logs, and it demonstrates precisely how decisive that data becomes once someone obtains it.

Libby Zion, 1984: questions a log now answers

In March 1984, Libby Zion, an 18-year-old college student, was admitted to New York Hospital with a fever and a flu-like illness. She died within hours of arriving. She had been taking phenelzine, a monoamine oxidase inhibitor, and she received meperidine during the admission. Her care was delivered largely by an intern and a junior resident, and no attending physician saw her in person that night.

What followed became a landmark in graduate medical education rather than a criminal matter. A grand jury declined to indict any individual and instead faulted a training system that left exhausted and thinly supervised residents responsible for an acutely ill patient. New York's Bell Commission followed with recommendations that became the state's resident work-hour rules, and the national accrediting body adopted comparable limits in 2003.

Look closely at which questions consumed the case. Who was contacted overnight, and at what time? What was actually communicated in that call? What information about the patient's existing medications was available, and to whom, at the moment a decision was made? In 1984 each of those could only be approached by asking people to reconstruct a single night from recollection. Today every one of them is a query against a log.

Betsy Lehman, 1994: the error that sat in the record

Betsy Lehman was a 39-year-old health reporter for the Boston Globe. In November 1994 she was receiving high-dose chemotherapy at the Dana-Farber Cancer Institute as part of a breast cancer treatment protocol. A fourfold overdose of cyclophosphamide was ordered, and she died on December 3, 1994.

The detail that turned the case into a national event was not the ordering error itself but how long it went unrecognized. The order was in the written record. It passed through the hands of multiple clinicians and pharmacists across several days of administration, and the overdose was not identified until a routine review of trial data months afterward, in early 1995.

The response reshaped American patient safety. Dana-Farber rebuilt its internal safeguards, the case became one of the most-cited arguments for computerized order entry and automated pharmacy checks, and Massachusetts later named its state patient-safety center after her.

The irony worth sitting with: the computerized ordering and verification systems adopted in response to errors like this one are the very systems that generate the audit data at issue in modern discovery. They were installed to make error visible. When a plaintiff later asks what they recorded, producing it is frequently called burdensome.

RaDonda Vaught, 2022: proof of how decisive system data is

In December 2017 a patient at Vanderbilt University Medical Center, Charlene Murphey, was prescribed a sedative before an imaging scan. RaDonda Vaught, the nurse who retrieved the medication, used an override function on the automated dispensing cabinet and withdrew vecuronium, a paralytic agent, rather than the intended sedative. Murphey died on December 27, 2017.

In March 2022 a Tennessee jury convicted Vaught of criminally negligent homicide and gross neglect of an impaired adult, and acquitted her of the more serious charge of reckless homicide. She was sentenced to three years of probation.

The prosecution drew sustained criticism from nursing organizations, which warned that criminalizing an individual clinician's error undermines the voluntary reporting that patient safety depends on. That criticism is well founded, and nothing here is directed at the nurse. The point of interest for litigators is the evidence.

The case against her rested substantially on machine-generated records: the dispensing cabinet's own record of the override and the withdrawal, alongside the electronic documentation surrounding administration. That material established a sequence of events with a precision no witness could have supplied from memory. It was obtained, it was admitted, and it was relied upon.

That is the asymmetry. When the subject was one nurse, this category of evidence proved obtainable and reliable enough to support a criminal conviction. When a civil plaintiff requests the same categories from an institution, the answer is frequently that the request is unduly burdensome, that the material is not relevant to the claims, or that an access report has already been produced and covers the ground. It does not cover the ground.

An access log shows who viewed a chart. It does not show what was entered, when it was saved as against when it was signed, what was altered afterward, or which safeguards were overridden and by whom. Those are different records answering different questions, and a production containing only the first can look complete while omitting everything that decides the case.

• Override and exception records, which capture the moments a safeguard was bypassed rather than followed.
• Action-level audit data showing creation, modification, and deletion, not merely access.
• Save-versus-sign timestamps, which expose entries authored well after the events they describe.
• Revision history and addenda, which distinguish a disclosed correction from a silent edit.

The Vaught prosecution is the demonstration, not the exception. Native system data was decisive there because native system data is decisive. The live question in any given matter is not whether the evidence exists but whether anyone requested it by function, in its native export format, and with enough specificity that a partial production is recognizable as partial.

What this means for a matter on your desk now

If a pending case turns on timing, on what was known at a particular moment, or on whether documentation is genuinely contemporaneous, the answer has probably already been recorded by the system. Whether it reaches the file depends almost entirely on how the request is written. Ask for the access log and the action-level audit trail separately rather than as one ambiguous phrase. Specify the native export rather than the printed chart or the release-of-information output, which omit audit data altogether. Request note revision history including save and sign timestamps. Ask the facility to identify its EMR, its version, and the audit reports it is technically capable of running, so that any claim of unavailability can be measured against documented capability.

Zion and Lehman were litigated in an era that had no better option than recollection. That constraint is gone. The system wrote down what happened while it was happening, and it is still there. The only remaining question is whether it gets asked for correctly, and early enough to matter.