Skip to content

What a Normal CGM Range Looks Like in Someone Without Diabetes

Oak
A steel laboratory column of clear gel with a wavering amber band of light held between two etched grooves, in a dark lab.

In people without diabetes, continuous glucose monitor (CGM) readings sit between 70 and 140 mg/dL (3.9 to 7.8 mmol/L) for roughly 95 to 98 percent of the day. The 24-hour mean is close to 99 mg/dL with a standard deviation around 15 mg/dL, which puts the coefficient of variation near 15 percent. Overnight values cluster in the 80s and low 90s. Post-meal peaks usually land between 110 and 140 mg/dL and return to baseline within two hours. Brief excursions above 140 after a large refined-carbohydrate meal are common in healthy people and are not by themselves a finding.

Those numbers are the target you should compare your own export against. The rest of this page covers where the distribution comes from, how much of your observed spread is sensor noise rather than physiology, and how to compute the metrics yourself from the raw file.

The reference distribution and its width

The widely cited healthy-adult CGM studies report a median time in the 70 to 140 mg/dL band of about 96 percent, with an interquartile range of roughly 93 to 98 percent. Mean sensor glucose lands at 98 to 99 mg/dL. The important part is the spread across individuals. The central estimate is not the important part. A healthy person at the 25th percentile spends about 100 minutes a day above 140 mg/dL, and that is still inside the normal distribution.

This matters because the tight-range metric was built for a different population. Time in tight range (TITR) is defined as 70 to 140 mg/dL. It was formalized as a treatment target for people with type 1 diabetes, where achieving even 50 percent TITR is demanding 1. Borrowing a diabetes target and applying it to yourself without knowing the healthy distribution leads to over-reading normal variation.

The same caution applies to the metrics themselves. CGM metrics were standardized for glycemic management, and their clinical thresholds were derived in populations with diabetes 2. Outside that context they are descriptive statistics about your own physiology. They are not a grading scale.

Sensor error is a large fraction of what you will see

Before interpreting a 138 mg/dL peak, account for how the number was produced. A factory-calibrated 10-day sensor has a mean absolute relative difference (MARD) of roughly 9 percent against a laboratory reference, with accuracy varying across the wear period 3. At a true glucose of 100 mg/dL, a 9 percent MARD means a typical reading lands somewhere in the low 90s to high 100s.

Two consequences follow. First, a single reading above 140 mg/dL tells you almost nothing. Second, your measured coefficient of variation is inflated by sensor noise. If the healthy CV is around 15 percent and your sensor contributes error of similar magnitude, a portion of your observed variability came from the electrochemistry. That portion did not come from you.

Accuracy is also worst where physiology is most interesting. Error tends to be larger in the hypoglycemic range and during rapid rate of change, because interstitial fluid lags plasma by five to fifteen minutes. Sensor performance also degrades in settings with poor perfusion or interfering substances, which is why CGM in critical care and hospital use carries explicit caveats about reliability 4 5.

Getting the raw data out

Two exports cover most of the market, and they differ in ways that will break naive parsing.

Dexcom Clarity exports a CSV with the columns Index, Timestamp (YYYY-MM-DDThh:mm:ss), and Event Type. It also has Event Subtype, Glucose Value (mg/dL), and several others. The first ten or so rows are device metadata with Event Type values like FirstName and Device. Glucose rows are Event Type == "EGV" at a 5-minute cadence. Out-of-range readings appear as the strings Low and High, not numbers, so pd.to_numeric will produce NaN unless you map them to 39 and 401 first.

Abbott LibreView exports a CSV with a header line above the real header, so read it with skiprows=1. Glucose lives in Historic Glucose mg/dL for the 15-minute background stream and Scan Glucose mg/dL for manual scans, split by a Record Type integer column. Do not concatenate the two columns without thought, since scans are non-uniformly sampled and will bias any time-weighted statistic toward moments you chose to look.

If you want programmatic access rather than manual export, the Dexcom developer API returns EGVs as JSON with systemTime, displayTime, value, and trend. Use systemTime for anything involving intervals and displayTime for anything involving meals and sleep, because the latter carries the user’s time zone changes.

Metrics worth computing yourself

Start with a time-weighted calculation rather than a simple mean of rows, since gaps during sensor replacement will otherwise skew results.

import pandas as pd, numpy as np

df = pd.read_csv("clarity.csv")
df = df[df["Event Type"] == "EGV"].copy()
g = df["Glucose Value (mg/dL)"].replace({"Low": 39, "High": 401})
df["glucose"] = pd.to_numeric(g, errors="coerce")
df["t"] = pd.to_datetime(df["Timestamp (YYYY-MM-DDThh:mm:ss)"])
df = df.dropna(subset=["glucose"]).set_index("t").sort_index()

# resample to a strict 5-min grid; do not interpolate gaps > 30 min
s = df["glucose"].resample("5min").mean()
s = s.interpolate(limit=6)

mean_g = s.mean()
cv = s.std() / mean_g * 100
tir_70_140 = ((s >= 70) & (s <= 140)).mean() * 100
above_140 = (s > 140).mean() * 100
gmi = 3.31 + 0.02392 * mean_g
print(mean_g, cv, tir_70_140, above_140, gmi)

Report the fraction of expected readings you captured. Below about 70 percent coverage over 14 days, your daily percentages are unstable. Fourteen days is the usual minimum for a stable estimate of mean and time-in-range.

For meal responses, the metric to compute is the incremental area under the curve over two hours from the first rise, with baseline taken as the mean of the 15 minutes before the meal. Peak height alone is noisy because it depends on a single sample. Repeat the same meal on three separate days before drawing any conclusion, since within-person day-to-day variation for an identical meal is substantial.

Failure modes that manufacture abnormal-looking data

Compression lows are the most common artifact. Lying on the sensor reduces local perfusion and produces a smooth 20 to 40 mg/dL drop lasting 30 to 90 minutes, usually overnight, with a sharp recovery on turning over. Flag any nocturnal nadir under 65 mg/dL that recovers without eating as suspect.

The first 12 to 24 hours after insertion often read low and drift as the wound response settles. We exclude the first 12 hours of every sensor session before computing summary statistics. Comparing two sensors worn simultaneously on the same person routinely shows mean differences of 5 to 10 mg/dL, which is enough to move your apparent mean glucose between sessions.

Fasting and meal timing dominate the daily curve more than most people expect. Shifting the entire eating window changes excursion patterns and overnight levels substantially, an effect documented clearly in CGM studies during Ramadan fasting 6. If you change your schedule mid-wear, analyze the periods separately.

What this can and cannot tell you

A CGM in a person without diabetes gives a high-resolution description of glucose behavior under real meals, sleep and exercise. It can show you which of your habitual meals produce two-hour excursions and whether your overnight baseline is stable. Studies in type 2 diabetes show that this kind of feedback changes food choices 7, and the same mechanism plausibly operates in anyone.

It cannot diagnose anything. Diabetes and prediabetes are defined by fasting plasma glucose, oral glucose tolerance testing and HbA1c measured in a laboratory. Sensor data does not define them. Your export may show a mean glucose above about 120 mg/dL, frequent readings above 180, or repeated verified values below 60 mg/dL. Any of those is a reason to see a physician for laboratory testing rather than a reason to change your diet based on the trace. Bring the CSV with you.

Questions people also ask

What is a normal CGM range for a non-diabetic?

Roughly 70 to 140 mg/dL for 95 to 98 percent of the day, with a 24-hour mean near 99 mg/dL and a coefficient of variation around 15 percent. Overnight values typically sit in the 80s to low 90s.

Is it normal to go above 140 mg/dL after a meal?

Yes. Healthy people commonly cross 140 mg/dL after a large, rapidly absorbed carbohydrate load, and the excursion usually resolves within two hours. Sustained time above 140 across many days and varied meals is the pattern that warrants laboratory testing.

Should I worry about a reading of 55 mg/dL overnight?

Check whether you were lying on the sensor. A smooth overnight dip with rapid recovery and no symptoms is most often a compression artifact rather than true hypoglycemia. Repeated low readings confirmed by a fingerstick meter should be discussed with a clinician.

How many days of CGM data do I need?

Fourteen days with at least 70 percent data capture is the standard minimum for stable estimates of mean glucose and time in range. Single days are too noisy to compare against reference distributions.

Does a CGM replace an HbA1c or a glucose tolerance test?

No. The glucose management indicator derived from CGM mean glucose is an estimate, and it can differ from measured HbA1c by a percentage point or more in an individual. Diagnosis rests on laboratory assays ordered and interpreted by a clinician.

Footnotes

  1. Roy W. Beck, Dan Raghinaru, Peter Calhoun, et al. A Comparison of Continuous Glucose Monitoring-Measured Time-in-Range 70–180 mg/dL Versus Time-in-Tight-Range 70–140 mg/dL. Diabetes Technology & Therapeutics, 2024. https://doi.org/10.1089/dia.2023.0380 ↩

  2. Tomasz Klupa, Leszek Czupryniak, Grzegorz Dzida, et al. Expanding the Role of Continuous Glucose Monitoring in Modern Diabetes Care Beyond Type 1 Disease. Diabetes Therapy, 2023. https://doi.org/10.1007/s13300-023-01431-3 ↩

  3. R. Paul Wadwa, Lori M. Laffel, Viral N. Shah, et al. Accuracy of a Factory-Calibrated, Real-Time Continuous Glucose Monitoring System During 10 Days of Use in Youth and Adults with Diabetes. Diabetes Technology & Therapeutics, 2018. https://doi.org/10.1089/dia.2018.0150 ↩

  4. James S. Krinsley, J. Geoffrey Chase, Jan Gunst, et al. Continuous glucose monitoring in the ICU: clinical considerations and consensus. Critical Care, 2017. https://doi.org/10.1186/s13054-017-1784-0 ↩

  5. Isaiah K. Mensah, Christopher W. Farnsworth. Continuous glucose monitoring devices in hospitalized pediatric patients. Critical Reviews in Clinical Laboratory Sciences, 2026. https://doi.org/10.1080/10408363.2026.2722682 ↩

  6. N. Lessan, Z. Hannoun, H. Hasan, et al. Glucose excursions and glycaemic control during Ramadan fasting in diabetic patients: Insights from continuous glucose monitoring (CGM). Diabetes & Metabolism, 2015. https://doi.org/10.1016/j.diabet.2014.11.004 ↩

  7. Thomas W. Martens, Holly J. Willis, Richard M. Bergenstal, et al. A Randomized Controlled Trial Using Continuous Glucose Monitoring to Guide Food Choices and Diabetes Self-Care in People with Type 2 Diabetes not Taking Insulin. Diabetes Technology & Therapeutics, 2025. https://doi.org/10.1089/dia.2024.0579 ↩

Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.