Methodology

Loop is an IRDAI-licensed insurance broker — Direct Broker (Life & General), registration IRDA/DB970/2022 — whose own account of its work covers primary care, chronic condition management and discounted diagnostics arranged through partner laboratories. The three datasets behind this report are by-products of that work: the blood panels from annual health checkups, a workforce health survey, and the hospital claims book. Each describes people covered by Loop policies — employees and the family members covered alongside them, in proportions that differ from one dataset to the next. None is joined to either of the others at the level of a person. What follows explains what each dataset counts, over what period, and under which definitions.

Three datasets, three different collection windows#

The three windows do not line up, because each dataset is dated on a different event: when a laboratory result was reported, when a patient was admitted, when someone pressed submit. The laboratory and claims years overlap almost entirely; the survey sits after both.

Biomarkers. These are annual health checkup panels run for Loop members. The unit is a report — one panel from one visit — and the file holds 816,003 individual result rows because a full panel covers dozens of analytes. Those rows resolve to 11,427 reports. A report is not a person: the analysis carries no person identifier at all, and no verified key exists in the source that would establish whether two reports belong to the same individual, so nothing in this report counts people. The base is not only the workforce, either. Every report carries a relationship label with two values — the covered employee, or a covered family member — and among the 9,585 reports with a haemoglobin result, 3,319 are covered family members.

Not every report carries every test. Each prevalence figure is computed over the reports that have a numeric result for that marker, so each marker has its own denominator: 7,088 reports carry an HbA1c result and 6,192 carry a B12 result, which is why the bases on the Blood Sugar and Vitamin B12 pages differ from each other and from 11,427. In total, 10,848 reports carry a numeric result for at least one of the markers used.

Where a report holds more than one result for the same marker, the analysis takes the first numeric value in file order rather than sorting by date. That situation arises on 9,409 report-and-marker pairs in the window; on 912 of them the two values actually differ, and all but 29 of those sit on a blood-count marker this report does not publish. The 29 are LDL, where taking the last row instead of the first would move the published LDL figure by a tenth of a point.

Survey. The survey collected 9,097 submissions, all of them between 6 March and 29 April 2026, covering sleep, stress, movement, nutrition, alcohol, nicotine, caffeine, supplements, preventive-care behaviour, GLP-1 awareness, where people get health information, and self-reported chronic conditions. The unit is a submitted response. A rolling twelve-month filter was applied to the submission timestamp; it retained every response in the file, so the filter selected the whole survey rather than a year of it, and the dates above are what the file actually contains rather than the width of the filter.

Repeat identifiers are kept rather than merged — 9,097 submissions carry 8,958 distinct respondent identifiers, and the source does not document what that identifier represents — so the figure counts submissions, not people. Every question has its own base: someone who answered about sleep did not necessarily answer about stress, and the eleven lifestyle questions' answered totals run from 6,968 to 7,614. All survey findings are what people said about themselves, which is the right instrument for behaviour and awareness and a weaker one for clinical status; the measured signal is in the biomarker dataset.

Claims. The claims dataset is Loop's broker book: 74,339 hospital claims admitted in the year to 26 December 2025. The public report does not disclose aggregate rupee totals. It reports relative shares and rates, and average amounts per claim. Claimed is what was submitted. Incurred is the amount recorded against the claim, including amounts on claims that are still open — it is not cash that reached anyone, and the gap between claimed and incurred is not the same quantity as the coded deductions. What a household actually paid is not in the data at all.

Each claim carries a relationship, an admission date, a free-text ailment description and an account tier. Of the three datasets, only claims separates covered family into layers: its relationship field groups into employee, spouse, child, parent or parent-in-law, sibling, and an "other" bucket for everything the grouping rule does not recognise. Pages that use those layers state the grouping they used; see Who the Policy Pays For.

Each filter runs from the file's latest valid date back 365 days, including both endpoints: 366 calendar dates. It selects the latest claims year; the biomarker and survey files fall entirely within their filters, with survey collection covering a shorter period.

The three datasets are not linked at the level of a person#

Before quoting anything in this report, know this: no verified key joins a person across these three files. A biomarker report cannot be matched to that person's survey answers or to their claims, and a shared city, sex or age label does not make two records the same individual. How far the three memberships overlap is unknown, and this report does not estimate it.

So when this report observes that the tested reports show a given rate of glucose dysfunction, that the survey shows a given rate of high stress, and that cardiac admissions account for a given share of the claims bill, those are three separate measurements of one broad audience, taken from three different sets of records over three different periods. They are not three views of the same records. The report does not say that the stressed respondents are the members with high HbA1c, or that either group is filing the cardiac claims, because the data cannot show it — and where a page places two such findings side by side, the reader should read them as two facts about an insured population, not as a chain.

What this report publishes#

Every figure in this report is an aggregate. No individual record, no name, no contact detail and no policy number appears in any published figure.

The claims and survey files are read with their identifying columns removed at load. On the claims side those are employee and claimant names, manager name, two email fields, the two named account-manager fields, a customer-system record identifier, the claim number and the policy number; the claims file carries no mobile or telephone column for the loader to remove. On the survey side they are the submission identifier, the respondent identifier, email and mobile. The laboratory file is never carried through whole: it holds a patient name and an address, and the analysis reads neither, keeping only a report identifier, age, sex, relationship, pincode and the numeric test values.

Sample sizes appear throughout for laboratory and survey findings, because a percentage without its denominator is not a finding. Claim counts for individual segments do not: per-segment claim volumes are commercially confidential. The public report may state the book-level total of 74,339 claims and may show relative shares/rates and average amounts per claim. It does not publish aggregate claim amounts.

What counted as elevated, low or deficient#

Laboratory prevalence on the risk pages uses the fixed cutoffs below. The laboratory's own reference range travels with each result in the source file, but the analysis uses the boundaries in this table. The operator is part of each definition: a result equal to the cutoff counts where the table reads ≥ or ≤, and does not where it reads > or <.

Where a marker has a sex-specific cutoff, the sex-specific one is what the risk pages use. A ratio between these rates compares results classified by different boundaries. It does not establish the difference in disease risk between men and women — see the gender divide. The liver measure counts a report as showing dysfunction if any of SGPT, SGOT or GGT is above its cutoff. The analysis performs no unit conversion: it treats each number as being in the expected unit for that marker, including where the unit label is missing. Each risk page names the cutoff behind its figures.

The survey side. The survey asked most of its questions as a single choice from a list, and a few as a set of tick-boxes where a respondent could choose several. That difference matters when reading any survey figure: the shares from a tick-box question overlap and do not add to 100%, and a "None of these" box is a selection someone made rather than proof that nothing applied. Some responses tick both the "None" box and an item on the same question: 32 on the supplement question, 53 on preventive actions, and 340 on the question about diagnosed chronic conditions. Where this report turns answers into a single flag, the answers are these:

Age groups are ten-year bands. Survey age tables are published for 20 to 59; the laboratory bands run wider, with an under-20 band and a 60-and-over band on the risk pages. The survey's age answer was typed in rather than chosen. Values outside 20 to 59 are either ordinary ages outside the published range or invalid entries. They are excluded from age tables but can enter a headline calculation when they meet that question's answer rules. The standard survey grouping function excludes groups with fewer than 50 answers, so a small city or income band may not appear in a cut even though its responses are in the headline total. Each page states the exact answer set and base it used.

A city here means a group of pincodes, not a workplace#

Every biomarker report carries a six-digit pincode. The city label is assigned from the first three digits, using the India Post allocation scheme, so a "city" in this report is a reporting area — a group of postal prefixes collected under one name. It is not a workplace, not a municipal boundary, not a single postal district, and not evidence of where anyone lives: what the file records is a pincode against a report.

The grouping is deliberately generous at the edges: Thane, Navi Mumbai, Kalyan and Vasai-Virar sit under Mumbai; Noida, Greater Noida, Ghaziabad, Gurgaon and Faridabad under Delhi NCR; Pimpri-Chinchwad under Pune. Beyond the five WHI cities the table also names Chennai, Kolkata, Ahmedabad and a number of other cities. A pincode outside the table takes its first-digit postal zone as a label instead, and in this export every report carried a pincode the mapping could read, so none was dropped for want of a city. The five cities the report profiles are not the whole sample: they account for 9,888 of the 11,427 laboratory reports, or 86.5%, and city findings are computed only on the reports that fall in the city concerned. The city pages state their own bases.

Biomarker cities use report pincodes; survey cities use self-reported residence and standardised aliases. Claims record hospital city, not patient residence. City pages combine separate laboratory and survey samples, with no person-level link.

What can and cannot be compared with WHI 2025#

WHI 2025 is the only previous edition, and every comparison in this report is made page by page against the figures that edition published. Direct comparisons use the same definition in both editions. Published rates with missing or inconsistent older definitions can provide context, with those limits stated.

Shared numeric cutoffs still need matching rules. HbA1c at 5.7% and 6.5%, fasting glucose at 100 and 126 mg/dL, HDL at 40 mg/dL for men and 50 for women, LDL at 100 mg/dL, triglycerides at 150 mg/dL, vitamin D at 20 and 30 ng/mL, and vitamin B12 at 200 and 300 pg/mL are stated numerically in both editions. Matching the number is not always the same as matching the rule: WHI 2025 states the triglyceride boundary inclusively on its lipid page and exclusively on another, so that comparison is anchored to 150 mg/dL rather than to a shared operator.

The liver comparison uses a matched SGPT measure. WHI 2025 published its liver finding as a single enzyme, SGPT, counted at or above 40 U/L: 16.4% of the 2,039 professionals it tested. Recomputed on that same boundary, this edition reads 18.7% of 7,809 reports, 2.3 points higher. That is the whole of the liver comparison. The narrower SGPT-only measure on the Liver page counts results strictly above 40 and reads 17.9%, and its broader measure counts SGPT, SGOT or GGT above their cutoffs — a different question, not compared with 2025.

Where definitions cannot be matched, the rates do not show change. WHI 2025 printed the same sex-specific ESR reference range this edition uses, but every elevated-inflammation rate it published was counted against a single boundary of 30 mm/hr, so those rates and this edition's are answers to different questions. The same test applies to survey questions: if changed answer options measure different behaviours, the shares need a common definition before they can be compared.

Where WHI 2025 gave no cutoff, as with haemoglobin and creatinine, this report compares the rates that edition published and does not claim the two editions used the same boundary, because the 2025 pages do not say what boundary they used.

The claims book has no previous edition. WHI 2025 published no claims analysis, so no cost figure in this report is a change from anything. Where the two editions can be set side by side, What Changed Since WHI 2025 collects the comparisons.

The samples also grew. WHI 2025 states a survey base of 3,437; this edition analyses 9,097 submitted responses. On the laboratory side, WHI 2025's lipid page describes 214,142 blood tests across 2,820 working professionals aged 22 to 65, against 816,003 results across 11,427 reports here. The larger samples describe more records. Their size alone does not establish that either edition represents the Indian workforce. The 2025 edition counted professionals, while this one counts reports and responses. Age bands are not generally comparable either: WHI 2025 used at least three different groupings across its pages, running 22–30 / 31–40 / 41–50 / 51–65 on most laboratory pages, generational bands of 22–27 / 28–42 / 43–58 on several lifestyle pages, and the same ten-year bands this edition uses on one.

City differences between editions reflect different samples and definitions, not population health trends. For example, Pune's HbA1c base changes from 384 reports to 2,562, and Delhi NCR's from 130 to 689.