Why Two Check-up Packages Can Show Different Item Counts

What we counted

Our sample is the 78 package URLs in the supplied Shanghai/Guangzhou catalogue inventory: 36 Shanghai pages and 42 Guangzhou pages. We fetched all 78 again on 13 October 2026. Every table’s text matched the supplied corpus after whitespace normalisation.

We used three structural categories:

• Column header: a row containing only heading cells, with more than one cell, such as the item/name/notes labels. Excluded.
• Section heading: a heading-only row with one cell spanning a group, such as liver function. Excluded.
• Publication entry: a table row containing a data cell. Counted once, however many organs, measurements or components its wording mentions.

The audit produced 298 column-header rows, 312 section-heading rows and 3,481 publication entries. Those figures describe this dataset and its HTML templates. The parser does not decide whether an entry is an independent analyte, an examination, a duplicate, or clinically appropriate.

The package-count CSV lists every URL and the three counts. The entry-ledger CSV preserves all 3,481 entry labels and their notes, so a reader can trace the arithmetic back to particular rows. The standard-library Python script checks the two CSV files against one another and reproduces the published totals. Save all three files in one folder and run python reproduce_counts.py. This verifies the CSV arithmetic; it does not independently re-fetch or classify the source HTML, and it does not produce a clinical test count.

Two complete catalogues, two different ways of writing a list

Here are all six published options at Renji East and all six at Clifford. These are publication entries, with headings removed and support entries retained. Package labels identify the source pages; they are not recommendations about which option an individual should select.

Scroll horizontally to see all columns.

The different totals do not measure a difference in medical coverage. The Men 1 pages show why. Renji prints complete blood count as one entry and does not name its components. Clifford prints 18 entries beneath its blood-count heading, numbered 24–41. Those entries include distribution plots and a source-listed red-cell-volume-fraction entry alongside haematocrit. The page expressly avoids inferring an additional analyte from the latter wording.

There is no sound conversion from Renji’s one line to Clifford’s 18 lines without further specification. Conversely, calling Clifford’s 18 lines 18 unique laboratory tests would go beyond what this audit establishes.

The same mismatch appears outside laboratory panels. Renji combines blood pressure, height, weight, BMI and waist/hip measurements in one entry. Clifford lists height, weight and blood pressure separately. That is a difference in the unit being counted, not proof that either package has the same scope as the other.

Four details that change the arithmetic

1. A heading can be mistaken for another examination

Clifford Men 1 contains 100 HTML table rows: one column header, 18 section headings and 81 entries. Counting every row would add 19 non-entry lines.

Repeated headings can also inflate a shorter list. Huadong Men 1 contains 20 table rows distributed across its tables. Five are repeated column headings, leaving 15 entries. An automated scrape that returns “20 rows” has not established “20 tests.”

2. A number in a panel name is not a named component list

Renji Men 1 names a four-component lipid panel, but the accompanying note says its individual components are unspecified. We record one published panel entry and preserve the stated four-component label. We do not silently substitute the four named lipid entries from Clifford’s list.

Huadong Men 1 illustrates another limitation: its blood-count label contains “15,” while a combined CEA/AFP/CA199 label occupies one row. Searching for the word “panel,” counting commas, or adding every number would produce different answers. None of those methods resolves what the hospital will actually perform.

3. Some counted entries are support items

Clifford Men 1 ends with disposable blood-collection supplies, a report and breakfast. The page explicitly labels these three as support items rather than additional medical examinations. Renji Men 1 includes refreshments as one of its 40 entries.

Removing the three Clifford support entries leaves 78 other published entries. That subtraction is valid arithmetic, but “78 clinical tests” is still an unsupported conclusion: the remaining rows mix assessments, grouped scope and source wording that has not been clinically deduplicated. The audit therefore keeps the inclusive count and explains the exceptions rather than publishing a corrected medical-test total.

4. An option or unresolved entry remains conditional

Renai Men 3 has one MRI entry offering a choice of head, cervical spine or lumbar spine. Our count assigns it one row. It does not add three scans because three regions are named. The separate MRI section heading adds nothing to the entry count.

Sun Yat-sen Memorial Women 2 retains two unconfirmed source-list entries with conflicting eligibility information. Both remain in a faithful publication-row count, but neither becomes a confirmed included examination. Likewise, entries carrying supplier eligibility conditions cannot be assumed to apply to every person using that package label.

What the linked PDFs confirmed

We downloaded the currently linked Men 1 PDFs for Renji East and Clifford on 13 October 2026. These are MyChinaCheckup’s published package documents, not independently obtained hospital orders.

Text and visual spot checks support the examples above: Renji retains grouped panels and refreshments; Clifford separates the blood-count entries and explicitly identifies the three support items. The check concerns those examples, not a claim that all 78 HTML pages have been reconciled with PDFs or that the documents are word-for-word identical. The table counts come from HTML, not PDF line breaks.

Both filenames contain 20261004, while their displayed footers say 10 October 2026. We therefore identify the checked version by its full linked URL and downloaded-file hash, rather than treating the filename as its last update date. No prices are compared in this audit.

How to use an item count without losing its meaning

When a quotation gives a headline total, ask for its counting rule and the dated list behind it. Does it count panels or named components? Are section headings, supplies or report services included? Does one imaging entry mean several included regions or a choice among them? Which rows still need confirmation?

Keep the original row and its scope note together. Where two lists use different units, leave the comparison unresolved until the provider supplies the missing definition. Do not turn a missing component name into a confirmed exclusion, and do not count an unexplained repeat as another independent test.

For the broader decision about examinations, language support, reports and the full arrangement, use How to Compare Health Check Packages in China. This audit supplies the narrower evidence needed to understand a published count.

Limits of this audit

This is a census of the supplied 78-page inventory, not of all hospitals or packages in either city. It measures MyChinaCheckup’s publication structure on one date. The same examination can appear in several packages and therefore several times in the dataset. Translation, grouping, eligibility and unresolved source wording remain material limits. We did not verify bookings, hospital availability, independent clinical test totals or medical suitability. No city, hospital or package quality ranking can be inferred from these numbers.