We order too many labs. I'll include myself in that. Routine panels drawn without a specific clinical question behind them, and suddenly the inbox is flooded with borderline values that are almost certainly insignificant — but still require a response. Every single one. And somewhere buried in all of that noise is the result that actually matters. A biopsy. A high-risk HPV. A final pathology report. Text-based results that don't auto-flag, don't trigger alerts, and just quietly land in a pile.
That's the problem I dig into in this week's Inefficiency Insights — why physician inboxes are set up to miss the things that matter most, and how I'm using AI right now to close the loop faster without the cognitive overhead.
I heard a story secondhand of a breast surgeon who did a lumpectomy. The intraop frozen pathology was clean without evidence of malignancy. The specimen still went to the main pathology lab, as it always does. The final read, though, came back days later as cancer. The result landed in the surgeon's inbox for their review. But it just sat there. Either the surgeon didn’t see it or something else. Those results sat there… nobody touched it… for a long time.
The point is that this wasn't a freak accident but a predictable failure. A result came back, landed somewhere it could be ignored, and got ignored. And when I heard the story, my first reaction wasn't to blame the surgeon (if you’re a physician or provider reading this, you’ve probably had some close calls like this with your swamped inbasket). It was to think about my own inbox.
I'm a resident. So my outpatient panel is somewhere around 70 to 100 patients. That's already enough to make result management a real task. Now imagine you're a PCP managing 500 to 1,000 outpatients. Or a surgeon with an even larger panel, who spends most of their day in the OR and not sitting at a workstation reviewing their inbasket. Their relationship with Epic looks nothing like mine.
And layered on top of the volume problem is one we created ourselves: we order too many labs. I include myself in this. In the outpatient setting, I've ordered routine panels without a specific clinical question driving them—”just checking in,” seeing what's there. What comes back is usually a sea of borderline values. Potassium slightly high. White count just below normal. Clinically insignificant, almost certainly. But each one still requires a response. Every. Single. One.
So the inbox fills up with noise. And somewhere in that noise is the result that actually matters. A biopsy. A high-risk HPV. A final pathology report. These don't trigger the same automated alerts that a critically low hemoglobin does. They're text-based, so they typically just land, quietly, in a pile.
Root Cause Analysis: 5 Whys
The 5 Whys process in root cause analysis involves repeatedly asking "Why?" five times to drill down into the root cause of a problem by exploring the cause-and-effect relationships underlying the issue.
The problem: Critical and time-sensitive patient results go unreviewed, leading to delayed or missed diagnoses.
- Why? Physicians receive a high volume of results with many flagged as abnormal, making it difficult to triage what truly needs urgent attention.
- Why? We over-order labs. Routine panels are drawn without specific clinical indication, flooding inboxes with results that need a response even when they're clinically meaningless.
- Why? Physicians don't have adequate administrative time built into their schedules to systematically review and close the loop on every result.
- Why? The majority of physician time is consumed by direct patient care, leaving little margin for the inbox work that follows.
- Why (root cause)? The clinical workflow treats result follow-up as an afterthought (a task layered on top of an already saturated schedule) with no reliable safety net to catch what falls through.
Impact Analysis
Impact analysis is the assessment of the potential consequences and effects that changes in one part of a system may have on other parts of the system or the whole.
- Patient: Delayed results mean delayed diagnoses and delayed treatment. In the case of malignancy, every week matters. Beyond outcomes, there's the psychological toll, since patients can now view their own lab results in MyChart or their EHR's patient portal before their physician has even reviewed them. They're sitting with unexplained findings, anxious, waiting for a message that may not come quickly enough.
- Clinician or Provider: Physicians are caught between two bad options: spend hours each day closing the loop on results, many of which are unremarkable, or risk missing the one that isn't. Both options carry a cost. The former contributes to burnout. The latter contributes to harm. At the end of the day—if you, the physician—ordered the test, it’s your responsibility to close the loop on results!
- System: Missed results generate downstream consequences such as additional workup, delayed treatment, longer hospital stays, legal exposure, and eroded patient trust. The cascade effect of a single missed pathology report can be significant, and the system has no built-in mechanism to prevent it from happening again.
Solution
A few years ago, I wrote about delays in communicating lab results with patients, highlighting the burden of manually reviewing and responding to every result, one by one. At the time, AI-assisted solutions were more theoretical than practical. This has, of course, changed.
Today, I'm using generative AI in my own workflow. When results come in, I can copy the last five values from the CBC or metabolic panel, paste them into Doximity Ask with a simple prompt…
Use the information provided to give a patient-friendly interpretation of results and what the next step(s) will be (I will mention what the next steps will be). Keep it succinct, only 2-3 sentences. Friendly. Middle school level.
…and within 10 seconds I have a draft patient message ready to send. I can add clinical context on top: follow up with GI, referral placed, repeat in three months. It's not perfect, but it's dramatically faster than typing every response from scratch, which is what I did as an intern.
See my demo video here 👇
Epic has started doing something similar natively by auto-generating responses for routine normal results. Some physicians like it. Others don't. But the direction is right.
The more important frontier, though, is what comes next.
For routine lab values, AI assistance is already feasible and increasingly available. The harder problem is text-based results: biopsy reports, imaging reads, pathology findings. These don't have numerical cutoffs. They don't auto-flag. An AI has to actually read the text, interpret it in clinical context, and determine whether something warrants urgent attention.
That's where the missed cancer story lives. Frozen path was negative. Final path came back positive. The result was text, not a flagged value. It landed in an inbox and got buried.
The solution here requires a few things working together. First, AI that can read the full body of an imaging report or pathology note, not just the impression section. Radiologists write impressions as summaries, and they sometimes omit findings that were clinically important to the ordering physician. The AI needs to be reading the whole thing. Second, persistent alerting for critical findings. If a result indicates malignancy and no response has been generated, the system should keep surfacing it (not once, not quietly) until the physician acknowledges it. Third, an AI co-pilot that tracks whether the loop was actually closed: was a message sent? Was the patient informed? Was a follow-up ordered? Essentially, this is the “Swiss Cheese Model” in action.
On top of that, these workflows need to be integrated into how each physician actually works, not bolted on as an external tool. The AI should know why the test was ordered, have access to the prior documentation, and be able to draft a contextually appropriate response. Providers can then interact with it—add treatment plans, adjust tone, include referral information—rather than starting from zero.
We also have to address the upstream problem: we order too many labs. Routine panels drawn without clinical indication create noise that drowns out signal. Tightening our ordering habits reduces inbox burden and makes it easier to recognize when something genuinely matters.
In summary, patient result follow-up is a system problem masquerading as an individual one. We've blamed physician inattention when the real issue is that we've built a workflow that generates enormous result volume, provides no reliable triage mechanism, and treats follow-up as the physician's personal responsibility (which is deserved) with no backup when they're overwhelmed. AI has the potential to close this gap, but only if it's integrated thoughtfully, with a specific focus on the text-based results that slip through every current safety net.
Read the original article here.
