12. Against stage, age and sex · Dr. Seok's Tumour-Infiltrating Lymphocytes, First Steps

Now we tie the infiltration rate to clinical information. The TCGA clinical table curated by UCSC Xena carries age, sex, stage and survival, and joins to our table on the patient barcode.

One rule first. Count each patient once. A patient with two slides entering twice has twice the influence on the result. So til.patients() keeps only the slide with the most tissue per patient.

5,201 slides become 4,750 patients.

Python Runs in your browser

import til

s = til.slides()
p = til.patients()
print("%s slides  ->  %s patients" % (format(len(s), ","), format(len(p), ",")))
print("patients with survival data: %s" % format(int(p.OS.notna().sum()), ","))
print("patients with a recorded stage: %s" % format(int(p.stage.notna().sum()), ","))

p = 2.6 × 10⁻⁶. Statistically, "infiltration differs by stage" is true.

Python Runs in your browser

import til

p = til.patients()
q = p[p.stage.isin(["I", "II", "III", "IV"])]

print("%-5s%7s%9s%8s%8s" % ("stage", "n", "median", "Q1", "Q3"))
for s, g in q.groupby("stage"):
    print("%-5s%7s%8.2f%%%7.2f%7.2f"
          % (s, format(len(g), ","), g.til_pct.median(),
             g.til_pct.quantile(.25), g.til_pct.quantile(.75)))

from scipy import stats
h, pv = stats.kruskal(*[g.til_pct.values for _, g in q.groupby("stage")])
print()
print("Kruskal-Wallis  H = %.2f   p = %.2e" % (h, pv))

But by how much? The gap between the highest stage (I, 6.11%) and the lowest (II, 4.77%) is 1.34 pp.

The measurement error we found in lesson 11 was 1.8 pp. The difference we found is smaller than the measurement error.

And the direction is odd too. It falls from I to II and rises again from II to III. This is not a tidy monotone story about "infiltration dropping as stage rises".

Click through stage, age and sex. The table below splits by cancer type.

Loading…

The impression from the pooled table scatters once the cancer types are separated.

Python Runs in your browser

import til

p = til.patients()
p = p[p.stage.isin(["I", "II", "III", "IV"])]

print("%-6s" % "type" + "".join("%8s" % ("st " + s) for s in ["I", "II", "III", "IV"]))
for t, g in p.groupby("tumor"):
    if len(g) < 80:
        continue
    line = "%-6s" % t
    for s in ["I", "II", "III", "IV"]:
        h = g[g.stage == s]
        line += "%8.1f" % h.til_pct.median() if len(h) >= 15 else "%8s" % "-"
    print(line)
print()
print("Split by cancer type, some rise with stage, some fall, some wander.")

🤔 Check yourself

How should you report 'p = 2.6 × 10⁻⁶ but a gap of 1.34 pp' for the stage difference?

🤔 Check yourself

Why keep only one slide per patient?