My job as a doctor is to take data and apply it to real people

Medicine is a little like cooking with a recipe written by a committee. The evidence tells you, “Bake at 350°F for 30 minutes.”
The patient in front of you says, “Cool, but my oven runs hot, I’m allergic to eggs, and I only have a toaster.”
My job is to take the best available datastudies, guidelines, lab values, risk scores, scans, and the occasional frantic Apple Watch alert
and translate it into care that actually fits a living, breathing human.

That translation is where the magic (and the chaos) lives. Data is clean. People are… not. People have jobs, fears, budgets, kids, insomnia,
cultures, cravings, traumas, and a remarkable ability to develop symptoms exclusively on weekends. So yes, I use numbers.
But I use them the way you use a GPS: as guidance, not as a command to drive into a lake just because the screen confidently suggested it.

Data is not the patient (and averages don’t wear hospital gowns)

Most medical evidence starts with groups. Hundreds, thousands, sometimes millions of people are studied to answer questions like:
“Does this drug lower blood pressure?” or “Does screening reduce cancer deaths?” The results are powerfulbut they’re still about populations.
Your patient is not a population. Your patient is one person with one body and a completely original relationship with caffeine.

Averages can be helpful, but they can also be misleading. If the “average” adult male is 5’9″, that doesn’t mean everyone should buy the same pants.
In clinical decision-making, the same logic applies: a treatment with clear benefits on average may be the wrong choice for a specific person
because of side effects, other conditions, medications, pregnancy, age, values, or plain old “I tried that once and it was a disaster.”

What the data really gives us

• Probabilities (risk of heart attack, chance a test is a false alarm)
• Tradeoffs (benefits vs. harms, symptom relief vs. side effects)
• Patterns (what tends to happen, not what must happen)
• Confidence (how sure we areand how sure we are that we’re not sure)

The data pipeline: from symptom to story

In real life, “data” doesn’t arrive as a neat spreadsheet. It arrives as a story:
“Doc, I’ve been tired for months.” Or: “My chest feels weird when I climb stairs.” Or: “My smartwatch says I’m dying.”
Turning that into usable information means building a pipelinehistory, exam, and targeted testingthen checking whether the pieces actually fit.

Step 1: The history (a.k.a. the human API)

The patient’s narrative is data. Onset, timing, triggers, what makes it better, what makes it worse, what they’re worried it is
these details shape the “pre-test probability,” even if nobody says that phrase out loud in the room because it sounds like an accounting term.
When a patient says, “This is not my usual headache,” that sentence can matter more than a dozen normal numbers.

Step 2: The exam (pattern recognition with hands and eyeballs)

Physical findings are data toosometimes messy, sometimes subtle, sometimes screamingly obvious (like a feverish person shivering under three hoodies).
The exam helps decide whether tests are needed, which tests, and how urgently.

Step 3: Testing (labs, imaging, and the seductive power of “normal”)

Tests are not truth. Tests are clues. A “normal” lab value might be normal for the general population and still wrong for this patient.
Reference ranges are averages from “healthy” groups, and many factors (age, sex, meds, hydration, timing) can shift results.
That’s why interpreting lab results always lives inside context, not in isolation.

The evidence toolkit: guidelines, trials, and the messy middle

Evidence-based medicine isn’t “do whatever a guideline says.” It’s more like: use the strongest evidence available, apply clinical judgment,
and respect patient preferences. Guidelines are incredibly useful because they summarize mountains of research and try to standardize good care.
But guidelines also assume a “typical” patientand typical patients are suspiciously rare in the wild.

Clinical trials: controlled answers to controlled questions

Randomized clinical trials (RCTs) are often the gold standard for figuring out whether an intervention works.
They’re designed to reduce bias by randomly assigning treatments and comparing outcomes. The upside is clarity.
The downside is that trial participants may not fully represent the patient in your room: different ages, different comorbidities,
different access to care, different life stressors, and a very different willingness to fill out 47 surveys.

Real-world evidence: what happens outside the lab coat bubble

Once treatments move into everyday practice, we learn morehow patients tolerate them, how adherence changes outcomes,
and which subgroups benefit most or least. Modern health systems increasingly use electronic health records (EHRs)
and clinical decision support to help scale evidence-based practice, but the goal is support, not autopilot.

Translating risk: numbers people can actually use

A big part of my job is risk communication: turning “relative risk reduction” into something a patient can feel in their bones.
If you tell someone, “This drug lowers your risk by 30%,” they’ll often hear, “So I’m basically immortal.” (Adorable, but no.)
Absolute risk is usually more honest and more useful.

A concrete example: the statin conversation

For cardiovascular prevention, clinicians often estimate a person’s 10-year risk of atherosclerotic cardiovascular disease (ASCVD)
using validated risk calculators. The point is not to label someone as “good” or “bad” at health; it’s to guide an individualized discussion:
how much benefit a statin might provide, what side effects are possible, and how the person feels about daily medication.

Two patients can have the same cholesterol number and different decisions. One might have diabetes, high blood pressure,
and a strong family historymaking the benefit of prevention bigger. Another might have a low overall risk and a history of muscle symptoms
with statinsmaking the tradeoff less appealing. The data gives probabilities. The patient supplies priorities.

Screening is risk too (just a different flavor)

Preventive screening is a perfect example of data meeting real people. Screening can reduce deaths for some cancers,
but it also carries harms: false positives, unnecessary biopsies, overdiagnosis, and treatments for problems that never would’ve caused trouble.
The best screening decisions often require individualized, shared decision-makingespecially when the “net benefit”
depends heavily on age, risk factors, and personal values.

When “normal” isn’t normal: interpreting lab results like a grown-up

Patients love the word “normal.” Clinicians do toobecause it’s comforting and fast. But “normal” can be a trap.
Reference ranges are not force fields that repel disease. Some people with significant illness can have “normal” results,
and some healthy people can drift slightly outside the range and be totally fine.

How I explain it in the room

I often say: “This number is one puzzle piece.” Then I connect it to symptoms, timing, medications, and what else we see.
A mildly abnormal thyroid test in a person with no symptoms might mean “recheck later.”
The same result in someone with weight changes, palpitations, and heat intolerance might mean “act now.”
Same data. Different human. Different plan.

Shared decision-making: two experts, one body

Here’s the underappreciated truth: patients are experts too. I’m the expert on medical evidence and pattern recognition.
They’re the expert on their lived experiencewhat symptoms feel like, what outcomes matter, what risks feel acceptable,
what “quality of life” means in their actual Tuesday.

Shared decision-making shows up everywhere: vaccination choices in certain scenarios, cancer screening decisions,
starting or stopping medications, or choosing between surgery and rehab. High-quality care often looks like a conversation,
not a command. The medical data frames the options; patient preferences help choose the path.

What shared decision-making sounds like

• “Here are the benefits we expect, and here are the harms we worry about.”
• “Here’s what the evidence says for someone with your risk profile.”
• “Tell me what you’re most hoping to avoidand what you’re most hoping to achieve.”
• “Let’s pick something we can actually do, not something that looks pretty on paper.”

Precision medicine: when data gets personal (in a good way)

Not all personalization is vibes. Some of it is biology. Genetics and pharmacogenetics can sometimes explain why a medication works beautifully
for one person and acts like a chaos gremlin in another. The goal isn’t to turn every clinic visit into a sci-fi movie,
but to use targeted tools when they genuinely change decisionslike selecting therapies, dosing, or assessing disease risk in certain contexts.

Even then, genetics isn’t destiny. Test results can raise or lower probabilities, not write the whole script.
Interpreting these results is another “data-to-person” translation job: what does this mean now, what doesn’t it mean,
and what decisions does it reasonably change?

The shadow side: bias, overuse, and diagnostic error

Data can help us, but it can also mislead usespecially when we treat tests as trophies (“Look! A CT scan!”) instead of tools.
Unnecessary testing can trigger a chain reaction: incidental findings, more scans, more procedures, more anxiety, and sometimes real harm.
Choosing tests and treatments wisely is part of evidence-based medicine too.

Overuse and “because we can” medicine

Many professional efforts encourage clinicians and patients to talk openly about whether a test or treatment is truly needed.
A good question is: “What problem are we trying to solve?” If the test result wouldn’t change the plan, we should think twice.
(This is also how you avoid discovering a “tiny harmless thing” that becomes a five-month saga of appointments.)

Diagnostic errors: when data is missing or misread

Diagnostic mistakes often come from a complex mix of factors: time pressure, fragmented records, cognitive bias,
and rare conditions masquerading as common ones. Better systems, better communication, and better decision support can help,
but the core remains the same: gather the right data, interpret it thoughtfully, and keep the patient’s story centered.

How I actually do it: a repeatable approach without becoming a robot

The goal is consistent thinking, not cookie-cutter care. Here’s a practical framework I use (and teach):

1) Start with the stakes

What are we worried about? What can’t we miss? A mild sore throat and a possible stroke do not get the same diagnostic workflow,
even if both patients are equally dramatic on the internet.

2) Estimate baseline likelihood

Based on the story and exam, how likely is the diagnosis before testing? This matters because tests behave differently
depending on how likely the disease is in the first place.

3) Choose the smallest helpful test set

“More data” is not always “better care.” The best tests are the ones most likely to change the plan.

4) Translate results into action

A number should lead somewhere: reassurance, follow-up, treatment, or additional evaluation. If it doesn’t, it’s just trivia with a copay.

5) Share the decision

When there are reasonable options, we talk. Medicine is full of “it depends,” and that’s not a weaknessit’s honesty.

Real-world experiences: where the data meets a face (and sometimes a sense of humor)

The phrase “apply data to real people” sounds tidy until you meet real people. Here are a few moments that live rent-free in my memory,
each one a reminder that numbers are only half the job.

1) The “perfect labs” patient who felt awful

A patient came in exhausted, foggy, and frustrated. Their basic labs were pristinean overachiever’s report card.
But the story didn’t match the spreadsheet. Digging deeper (sleep, mood, stress, medication timing) revealed severe insomnia and anxiety,
plus a stimulant taken too late in the day. The “data” that solved it wasn’t a lab value; it was the pattern of their life.
We adjusted the routine, treated the underlying issue, and the improvement was dramatic. Clean labs, real sufferingboth true at once.

2) The borderline number that mattered because of the person

Another patient had a borderline blood pressure reading that most people would ignore. But they had a family history of early stroke,
limited access to follow-up, and a job that made lifestyle changes realistically tough. We used home readings, talked about absolute risk,
started a low-dose medication, and built a plan they could maintain. The guideline didn’t “force” the decisionthe context shaped it.

3) The screening conversation that turned into a values conversation

A patient asked, “Should I get this cancer screening test?” The data offered pros and cons, not a single correct answer.
When we talked about what they feared mostdying young versus living with complications from unnecessary proceduresthe choice became clear
for them. The best part? They left feeling empowered, not scolded, because the plan matched their values, not my assumptions.

4) The smartwatch panic

Wearables are amazing and occasionally unhinged. A patient showed me an alert suggesting an arrhythmia.
Instead of dismissing it or declaring it gospel, we treated it like any other data point: symptoms, risk factors, targeted testing.
The conclusion was reassuringbut we also used the moment to talk about when to seek help and how to interpret alerts without spiraling.
Technology gave us a clue; clinical judgment gave it meaning.

5) The “I don’t care about the number” moment

I once explained a medication’s risk reduction in careful, friendly terms. The patient listened and said,
“Doc, I hear you. But I care more about not feeling nauseated every day than I care about that percentage.”
That wasn’t noncompliance. That was clarity. We changed the plan. The data didn’t lose; it just stopped pretending it was the only input.

Experiences like these teach the same lesson repeatedly: evidence-based medicine is not evidence-only medicine.
The data helps us see the landscape; the person tells us where they’re willing to walk. And when we do it well,
patients don’t just get “a recommendation.” They get a plan that respects both science and reality.

Conclusion

My job as a doctor is to translate: from population data to individual care, from lab ranges to lived experience,
from clinical guidelines to shared decisions that fit a person’s body and life. The best outcomes come from pairing solid evidence
with thoughtful interpretation and honest conversation. Data mattersdeeply. But it becomes medicine only when it’s applied with judgment,
humility, and a focus on the human in front of us.