Address potential mixture of Cushing's etiologies

[davebridges]

Gap 8: Case definition relies on ICD codes without etiology validation — cohort may mix pituitary, adrenal, ectopic, and possibly iatrogenic Cushing's

Labels: reviewer-response case-definition data-extraction priority-medium

Reviewer summary

Case identification uses ICD codes E24.0 and 255.0, which capture Cushing's syndrome/disease broadly rather than pituitary-dependent Cushing's disease specifically. No chart review or biochemical validation confirms pituitary etiology. Reviewer cites Zhou et al. 2019 showing ICD code validity concerns for Cushing's etiology classification. Proposes either (Option 1) a structured-code enrichment algorithm with tiered tiered classification using pituitary surgery CPT codes, pathology terms, IPSS codes, and steroid-exposure exclusion; or (Option 2) PHI-limited chart adjudication by two endocrinologists on n≈200 cases.

My critique of the critique

Legitimate:

• ICD E24.0 is "Cushing's syndrome" (generic), not "Cushing's disease" (pituitary-specific). The paper's title and prose use "Cushing's disease" throughout, overstating the specificity of the case definition.
• ICD 255.0 (legacy) was similarly broad.
• Patients with long-term systemic glucocorticoid exposure can be incidentally coded with Cushing's syndrome ICDs (iatrogenic Cushing's), which would contaminate an "endogenous hypercortisolism" cohort.
• The reviewer's cited Zhou et al. validation paper is directly relevant and is already cited in the manuscript — worth acknowledging what that paper actually found about etiology PPV rather than just citing it.

Overstated or misdirected:

• Option 2 (chart adjudication by two endocrinologists) requires IRB modification, PHI access, reviewer recruitment, and adjudication protocols. This is the design of a validation study, not a sensitivity analysis. Skip entirely and document in the response letter why.
• Three-tier (Tier 1/2/3) classification adds complexity for limited analytic gain. A binary "enriched probable-pituitary" subset vs full cohort answers the same question more cleanly.
• Etiology distinction may matter less than the reviewer assumes. Pituitary, adrenal, and ectopic Cushing's all produce the same endogenous hypercortisolism. For the paper's metabolic outcomes (glucose, liver enzymes, BP), the mechanism is cortisol excess irrespective of source. Where etiology matters is for disease course, treatment response, and specific treatment approach — which this paper doesn't study. Worth naming this explicitly in the response: the biology claims survive a broader "endogenous Cushing's" case definition.

What's actually fixable:

• Honest relabeling of the case definition and selective prose edits in the manuscript are ~80% of the response. The paper should describe the cohort as what it actually is.
• Enriched subset with pituitary surgery CPT as a secondary sensitivity.
• Glucocorticoid-exposure exclusion to rule out iatrogenic contamination.

What needs to change

A. Manuscript relabeling (primary response)

This is the most important change in this gap. Three edits:

1. Methods — describe the case definition accurately:

   Replace: "To identify patients with Cushing's disease, we evaluated electronic health records from participants at Michigan Medicine between January 1st 2000-2025 searching for diagnoses of Cushing's disease (including ICD Codes E24.0, and 255.0, validated in 9)."

   with: "To identify patients with endogenous hypercortisolism, we evaluated electronic health records from participants at Michigan Medicine between January 1, 2000 and 2025 for ICD-10 E24.0 (Cushing's syndrome) and ICD-9 255.0 (Cushing's syndrome and related conditions). These codes capture Cushing's syndrome of any endogenous etiology (pituitary, adrenal, ectopic) and may occasionally include patients with exogenous glucocorticoid exposure. Code validity has been previously assessed [ref 9 — Zhou et al.], which found [summarize their PPV findings]. Because metabolic consequences of hypercortisolism are expected to be broadly similar across endogenous etiologies, we did not require etiology-specific confirmation for the primary analysis; enrichment for probable pituitary-dependent disease is reported as a sensitivity analysis."

2. Results — adjust the count language:

   Replace: "We found 2,893 unique diagnoses of Cushing's disease."

   with: "We identified 2,893 unique patients with ICD codes for Cushing's syndrome or Cushing's disease."

3. Discussion / Limitations — add an explicit paragraph:

   "Our case definition relied on ICD codes without local chart or biochemical validation of pituitary etiology. The cohort likely includes a mix of pituitary-dependent Cushing's disease, adrenal Cushing's syndrome, and ectopic ACTH-dependent hypercortisolism, with possible inclusion of iatrogenic cases. Because the metabolic consequences of hypercortisolism (hyperglycemia, hepatic steatosis, hypertension) arise from cortisol excess regardless of source, the paper's interaction findings should generalize across endogenous etiologies; however, etiology-specific differences in disease course and treatment timing could introduce heterogeneity. We report sensitivity analyses restricted to patients with documented pituitary surgery (Table 4) to test robustness in a probable-pituitary-Cushing's subset."

4. Consider title change. "Differential Metabolic Signatures of Cushing's Disease Patients…" → "Differential Metabolic Signatures of Patients with Endogenous Hypercortisolism…" (or similar). This is a decision for the author team, not a mandate from the review, but the reviewer's concern is essentially about this framing mismatch. Worth raising among co-authors.

B. Enriched pituitary-Cushing's subset via CPT codes

Pituitary surgery CPT codes as positive enrichment markers:

• 61546 — Craniotomy for hypophysectomy or excision of pituitary tumor, intracranial approach
• 61548 — Hypophysectomy or excision of pituitary tumor, transnasal or transseptal approach
• 62165 — Neuroendoscopy, intracranial; with excision of pituitary tumor, transnasal or transseptal approach
• 61140 — Biopsy of brain lesion (used for pituitary biopsy in some institutional coding)
• 61575, 61576 — Transoral craniectomy approaches sometimes used for pituitary

library(dplyr)

pituitary_cpt <- c("61546", "61548", "62165", "61140", "61575", "61576")

pituitary_surgery_flag <- procedures |>
  filter(cpt_code %in% pituitary_cpt) |>
  inner_join(cases |> select(DeID_PatientID, cushings_procedure),
             by = "DeID_PatientID") |>
  filter(procedure_date <= cushings_procedure + days(30)) |>  # allow small coding lag
  distinct(DeID_PatientID) |>
  mutate(pituitary_surgery = TRUE)

cases_enriched <- cases |>
  left_join(pituitary_surgery_flag, by = "DeID_PatientID") |>
  mutate(pituitary_surgery = replace_na(pituitary_surgery, FALSE))

Report: n and % of cases with documented pituitary surgery. This is your probable-pituitary subset.

C. Exclude iatrogenic / chronic steroid exposure

Patients with ≥30 days of systemic glucocorticoid prescriptions in the 6 months before the Cushing's ICD code are candidates for iatrogenic Cushing's rather than endogenous hypercortisolism.

systemic_steroids <- c(
  "prednisone", "prednisolone", "methylprednisolone", "dexamethasone",
  "hydrocortisone", "triamcinolone", "betamethasone"
)

# Skip topical/inhaled/intranasal preparations — these don't typically cause
# systemic hypercortisolism at therapeutic doses.

chronic_steroid_exposure <- medications |>
  mutate(med_lc = tolower(med_name)) |>
  filter(
    str_detect(med_lc, str_c(systemic_steroids, collapse = "|")),
    !str_detect(med_lc, "topical|inhaled|nasal|ophthalmic|otic|cream|ointment|drops")
  ) |>
  inner_join(cases |> select(DeID_PatientID, first_cushings_dx_date),
             by = "DeID_PatientID") |>
  filter(start_date >= first_cushings_dx_date - days(180),
         start_date < first_cushings_dx_date) |>
  mutate(days_supply = pmin(as.numeric(end_date - start_date), 180)) |>
  group_by(DeID_PatientID) |>
  summarise(total_steroid_days = sum(days_supply, na.rm = TRUE), .groups = "drop") |>
  filter(total_steroid_days >= 30) |>
  mutate(chronic_steroid = TRUE)

cases_enriched <- cases_enriched |>
  left_join(chronic_steroid_exposure |> select(DeID_PatientID, chronic_steroid),
            by = "DeID_PatientID") |>
  mutate(chronic_steroid = replace_na(chronic_steroid, FALSE),
         probable_endogenous = !chronic_steroid,
         probable_pituitary  = pituitary_surgery & !chronic_steroid)

D. Re-estimate interactions in the enriched subsets

Two sensitivity rows for Table 4:

1. "Restricted to probable endogenous Cushing's" — excludes patients with ≥30 days pre-diagnosis systemic glucocorticoids.
2. "Restricted to probable pituitary Cushing's" — CPT-confirmed pituitary surgery AND no chronic steroid exposure.

# Primary analysis code, restricted subset
lm.alt.pituitary <- lm(
  value ~ GenderCode + RaceEthnicity + Cushings * Obesity,
  data = alt.data |>
    filter(Cushings == 0 | (Cushings == 1 & probable_pituitary))
)

Report n per stratum for each sensitivity row. The probable-pituitary subset will be smaller than the primary cohort; expect wider CIs.

E. Descriptive — how much does this actually change?

Report as a supplementary table:

• Overall Cushing's n: 365 (per manuscript)
• Pituitary surgery-confirmed: [N, %]
• Chronic steroid exposure (probable iatrogenic): [N, %]
• Probable endogenous: [N, %]
• Probable pituitary: [N, %]

If pituitary surgery confirmation covers >80% of the cohort, the case definition concern is mostly empirical noise. If it covers <50%, it's a more meaningful limitation and the relabeling edits in (A) become more important.

F. Optional — Bayesian subset analysis

For the probable-pituitary subset (likely smaller n), Bayesian regression with weakly informative priors is useful for stable inference. Same brms infrastructure as Gaps 4, 5, 7.

Acceptance criteria

• Pituitary surgery CPT flag extracted.
• Chronic systemic glucocorticoid exposure flag extracted.
• Supplementary table: composition of the cohort (pituitary-confirmed, endogenous-probable, steroid-excluded).
• Table 4: two new rows — "Restricted to probable endogenous Cushing's" and "Restricted to probable pituitary Cushing's" — for all primary interactions.
• Methods paragraph on case definition accurately describes the ICD-based ascertainment.
• Limitations paragraph added acknowledging etiology heterogeneity.
• Results count language updated ("patients with ICD codes for Cushing's syndrome or disease").
• Title and abstract language reviewed for "Cushing's disease" vs "endogenous hypercortisolism" — flag for co-author discussion if a change is warranted.
• Optional: Bayesian parallel for the smaller pituitary-restricted subset.
• Skip chart adjudication (Reviewer Option 2). Document in the response letter: "Chart adjudication requires IRB modification, PHI access, and a validation study design distinct from the present analysis. The pituitary-surgery-confirmed subset provides a structural-data proxy that tests robustness of the primary findings in a higher-specificity cohort."
• Skip three-tier classification. Document: "A binary enriched-cohort comparison (pituitary-confirmed vs full cohort) provides equivalent information with greater interpretability."

Notes / open questions

• Zhou et al. 2019 PPV findings. The manuscript cites this paper but doesn't report what it found. Worth summarizing: the paper examined PPV of discharge ICD codes for Cushing's syndrome etiology at a Chinese referral center. Incorporate a sentence on their numbers in the response letter so we're engaging with what they actually concluded, not just citing the existence of validation literature.
• "Treatment date" filter. The current Methods require a documented treatment date. For many Cushing's cases, this treatment is pituitary surgery — implicitly enriching for pituitary cases already. Worth verifying how many treatment-date cases overlap with the pituitary-CPT flag. If overlap is high, the current filter is already doing most of the work the enrichment would add.
• Adrenal and ectopic Cushing's are fundamentally the same disease biology (hypercortisolism). The paper's biology claims survive even if the cohort is a mix. The only time this mix matters analytically is if etiology correlates with obesity status (e.g., if adrenal Cushing's is more common in the obese stratum, that could confound). Descriptive data on etiology distribution by BMI stratum within the pituitary-confirmed vs unconfirmed subsets would address this.
• IRB scope. The current IRB (HUM00247814) was adjudicated "not regulated" because it uses deidentified data. Chart adjudication would require a new submission — another reason to skip Option 2.

References

1. Reviewer ref 1 (ICD validity for Cushing's etiology) — likely Zhou et al. 2019, already cited as ref 9 in the manuscript. Engage with their actual findings.
2. Administrative data misclassification in endocrine disorders (Reviewer ref 2).
3. Ditto (Reviewer ref 3).

(Locate full cites before response letter.)