From the clinic floor: what Grace actually does

The trial average is rarely the person in front of you

Twenty years in BWC paediatric T1D plus the years before that in adult diabetes. A school-age T1D with a recently diagnosed cousin and a parent who works shifts will not get the trial-cohort closed-loop outcome from the closed-loop pump, because the trial cohort had two parents, both at home, both motivated enough for 17 trial visits. The pump is doing what it is designed to do; the cohort is not what the marketing implied. The conversation in clinic is not “your closed-loop is not working”; it is “the closed-loop is working as it can given the environment; what can we change in the environment?”.

“Very few of us will behave exactly like the point average on the graph: some will be more susceptible to benefit and some more susceptible to harm from a particular intervention.”

Source: Greenhalgh, How to Read a Paper, 7th edition (2025), Chapter 16. The clinical-statistics literature names this individual-vs-mean problem; the clinic shows it every appointment. Both halves of the same fact.

Carb counting perfectionism is the most common form of fragility in T1D

A person who counts to the gram on every meal is the person whose system collapses when they eat out, when the recipe changes, when the carb estimate is twenty grams off. Per Taleb, “better to be roughly right than precisely wrong”; per the clinic, the family that estimates to the nearest ten grams and bolus-corrects from the CGM trend has better long-arc time-in-range than the family that hits the spreadsheet on every plate. The intuition is anti-fragility: the rough estimate strengthens under the stress of unfamiliar food; the spreadsheet shatters.

Time in range above 60% is a population marker, not a personal target

The 70% threshold from the consensus literature came from population-mean correlations with retinopathy and nephropathy outcomes. The relationship between a particular person’s TIR and their long-arc complications is not 1:1; it is one input among several (HbA1c level itself, variability, severe hypo count, weight, blood pressure, lipids, lived stress). Quoting “you need 70%” to a person on 56% who is otherwise stable is poor clinical practice; quoting it to a person on 32% who has severe hypo every week is the right conversation. The number alone is not the prescription.

The person who has lived with T1D longest knows the most

The clinic conversation works when the clinician treats the person with T1D as the person with the deepest expertise about that person’s diabetes, and the clinician brings the population-level knowledge. The opposite framing (the clinician knows; the person carries the diagnosis) produces worse outcomes by every measure we have. Goldacre’s anti-IYI critique applies here: the clinician who acts as if the trial average overrules the person’s lived experience is the clinician who is, structurally, the IYI of the consulting room.

Grace’s 5-paper synthesis

Grace reads roughly five papers per topic at the John tier. The 5-papers approach is Via Negativa in operation: the marginal information from paper six onwards is usually noise; the synthesis from five well-chosen papers is usually what the reader needs. The five always include:

1. The systematic review or meta-analysis at the top of the topic, if one exists and is current within five years. If two SRs disagree, both. If none exists, the senior consensus statement (ADA, EASD, NICE, ISPAD, IDF) fills the slot.
2. The most-cited pivotal RCT on the topic, with the CONSORT report and the trial-registration entry.
3. A real-world-evidence cohort that quantifies the gap between trial and lived outcome. This is the slot most often missing from clinical guidelines and most often load-bearing for the conversation in the clinic.
4. A mechanism paper that explains the why. (Twin-cycle hypothesis for T2 remission, IOB modelling for AID, etc.)
5. A dissenting voice or limitations paper that names what the consensus might have wrong. (Wieten 2018 on EBM critique, Goldacre on pharma bias, Greenhalgh on EBM’s blind spots, Ioannidis on false-positive prevalence.)

“The production of a diamond at the bottom of a plot is an exciting moment for many authors, but results of meta-analyses can be very misleading if suitable attention has not been given to formulating the review question.”

Source: Cochrane Handbook v6.5 (2024), Chapter 10, Section 10.1. The diamond looks definitive on the page; the question it answers may not be the question the learner is asking. Part 6 turns the synthesis into the auditor’s checklist you can run on any single paper.