I keep a small black notebook in the top drawer of my clinic desk. Its pages contain dates and 2 columns: “What I Predicted” and “What Happened.” I rarely show it to anyone. It began as a record of my accuracy. Over time, it became something else entirely.

On the day I met a woman with metastatic pancreatic cancer, I opened a fresh page. After reviewing the scans with her and her husband, I explained what the numbers suggested. “With treatment, we may be looking at six to nine months. Perhaps longer if things go well.” She nodded, then asked whether she would be able to attend her grandchildren’s school play. I remember pausing. She was not really asking me to predict her death. She wanted to know whether there was still time for something that mattered to her. “I think that’s possible,” I said carefully. She started chemotherapy the following week.

More than a year later, she walked into my clinic smiling. The cancer was still there, but unexpectedly controlled. She hugged me before leaving and said, “Thank you for giving me time.” After the door closed, I opened the notebook. My prediction stared back at me – confident, specific, wrong. In the second column, I wrote: Beautifully, mercifully wrong.

For years, I had seen my responsibility as answering questions about prognosis as accurately as possible. I tried to distill complexity into probability and offer families a number around which they could plan. The notebook was supposed to hold me accountable to that responsibility, but its pages began to tell a different story.

Another woman had metastatic breast cancer and was deteriorating quickly. I told her gently that I worried her time might be measured in weeks or, perhaps, a few months. She listened and then looked directly at me. “Doctor,” she said, “I’ve buried two husbands. I decide when I’m done.” She lived well beyond the few months I had anticipated. Near the end of her life, I visited her at home. She had spent one of her last good days in her garden, tending her plants.

Then there was a man with metastatic lung cancer and an actionable molecular alteration. I spoke to him and his wife about targeted therapy and the possibility of prolonged disease control. This time, I allowed myself to imagine years. Three months later, the treatment had failed. Soon afterward, I sat with his wife in the same consultation room where we had discussed the possibility of a long response. She tried to understand how a treatment that had offered so much hope had given them so little time. “We thought we had longer,” she said.

I had thought so too.

At first, these seemed like separate errors, contained within separate pages of the notebook. When patients lived longer than I predicted, I felt relieved, but the relief carried discomfort. Had I narrowed their horizon too early? Had a number offered in the name of honesty changed how they imagined the time ahead? When patients died sooner, the discomfort took another form. Had my optimism encouraged plans they would never reach? Had it delayed conversations their families needed to have?

More troubling was something I did not recognize at first. My language had begun to change. After several patients deteriorated sooner than expected, I became more cautious. My estimates shortened, and I found myself preparing families for the possibility that time might be less than we hoped. Then someone would do unexpectedly well, and my next conversation would contain more room for response and possibility.

I assumed each prognostic assessment began afresh, informed by the disease, the treatment options, and the available evidence. Slowly, I began to understand that I was carrying those who had come before into every new conversation. The woman who had lived far longer than I expected pulled me toward optimism. The man who died within months pulled me toward caution. The woman who challenged my estimate reminded me how easily a prediction could begin to sound like a verdict.

I thought the notebook was recording my mistakes. I had not realized that my mistakes were also recording themselves in me.

One afternoon made this impossible to ignore. I had recently seen the woman with pancreatic cancer again, still doing remarkably well. Around the same time, I found myself thinking about the man whose targeted therapy had failed so quickly and the time his family had expected but never received. Then another patient with advanced cancer sat across from me. Her daughter asked the question families always ask: “How long does she have?”

I had been asked that question hundreds of times. I looked at the scans and then at the woman sitting in front of me. For the first time, I was acutely aware not only of the evidence I was considering, but also of the experiences I had brought into the room. The problem was no longer simply whether my estimate would prove right or wrong. Every prognostic error had changed me, and unless I recognized that change, another person’s unexpected survival or early death could quietly shape the answer I was about to give.

“I can give you a range,” I said. “I can tell you what usually happens. But I have been wrong before, in both directions.” I explained what I thought was most likely, what a shorter course might look like, and what we might hope for if treatment worked better than expected. We talked about what mattered to her if time became short and what she still hoped to do if it did not. We held both possibilities in the room.

We also discussed involving palliative care early. Earlier in my career, I had thought of palliative care as something that entered when treatment options were disappearing. I see its role more broadly now. Palliative care does not demand certainty. It gives patients and families a way to live while the timeline remains unclear and creates room to revisit what matters as the illness changes.

That conversation changed the way I understood the notebook. It was no longer simply a scorecard. It had become a reminder to examine what I might be carrying into the room. Was I being cautious because of the person sitting in front of me, or because someone else had died sooner than I expected? Was I offering hope because the evidence supported it, or because someone had surprised me by living far longer than I imagined?

I still give prognoses. Patients deserve more than an evasive “I don’t know.” But my approach has changed. I use ranges more often and distinguish what I expect from what I know. I talk about the most likely course while making room for better and worse possibilities, and I revisit the conversation as the illness evolves. The challenge is not to eliminate uncertainty but to speak honestly within it.

Once spoken, a prognosis enters a family’s life. People make plans around what we tell them. They decide what can wait, which conversations need to happen now, and which hopes they can still afford to hold. Even a statistically reasonable estimate can acquire a certainty we never intended to give it. Yet silence is not safer. Our responsibility is to help patients prepare for the future without pretending we know precisely what that future will be.

Recently, after meeting another patient with advanced cancer, I returned to my desk and opened the notebook to a fresh page. For a moment, I considered what to write in the prediction column. Then I wrote: I don’t know yet.

I left the second column empty.

There was still time for the patient to tell us what happened.