Methodology

WHI 2026 is built on three datasets Loop holds as an IRDAI-licensed broker that also runs diagnostics, primary care, and chronic disease management: annual-checkup biomarkers, a workforce health survey, and the hospital claims book. Each describes the same population of urban working professionals and their covered dependents. None of the three is joined to either of the others at the level of an individual person. This page sets out where each dataset comes from, the windows and thresholds applied to it, how geography was assigned, how the year-on-year comparison was made, and where the data does not support the weight a reader might want to put on it.

The three datasets#

Each dataset is windowed to its own latest available twelve months. The windows do not align exactly, because each dataset is dated on a different event — when a lab result was reported, when a patient was admitted, when a survey was submitted.

Biomarkers. These are real annual-health-checkup blood panels run on Loop members over the twelve months ending December 2025. The dataset holds 816,003 individual result rows — each row a single parameter on a single panel — which resolve to 11,427 unique reports once grouped to the person-visit level. A report is one person's checkup; the row count is high because a full panel covers dozens of analytes. The panels cover glucose (HbA1c, fasting and random glucose), the lipid profile, haemoglobin, vitamin D, vitamin B12, the liver enzymes, kidney markers, uric acid, and inflammation markers. Prevalence figures in the Risk Signals pages are computed at the report level, not the row level.

Survey. The survey ran primarily across March and April 2026 and collected 9,097 responses to roughly 126 questions spanning sleep, stress, movement, nutrition, substance use, caffeine, preventive-care behaviour, GLP-1 awareness, and chronic-condition self-report. All survey findings are self-reported. They capture what respondents say about their behaviour and health, which is the right instrument for behaviour and awareness but a weaker one for clinical status — the biomarker dataset carries the measured signal.

Claims. The claims dataset is Loop's broker book over the twelve months ending December 2025: 74,339 hospital claims, ₹320.6 Cr incurred against ₹470.2 Cr claimed, with ₹98.1 Cr deducted. Each claim carries a relationship (employee, spouse, child, parent or parent-in-law), an admission date, an ailment description, an account tier, and the cashless-versus-reimbursement route. Claims is the only one of the three datasets that separates dependents into spouse, child, and parent layers.

The datasets are not linked at the individual level#

The single most important constraint on reading this report: the three datasets are not joined at the person level. A biomarker report cannot be matched to that same person's survey response or their claims. The datasets describe one broad audience universe — employees covered by Loop policies, plus their spouses, parents, and children — but they are three separate windows onto it, not three views of the same individuals.

Every statement that crosses datasets in this report is therefore an audience-level pattern. When a page observes that the same workforce showing 38.7% glucose dysfunction in the biomarkers also reports high stress in the survey and drives cardiac claims in the broker book, that is three parallel measurements of one population. It is not a claim that the stressed survey respondents are the dysfunctional-glucose members, nor that either group is the one filing the cardiac claims. The report does not assert member-level causal chains, because the data cannot support them.

Identifiers are stripped at load#

Identifier columns are removed when each file is read and never reach the analysis layer or any output. The stripped fields include patient name, address, employee names, claim numbers, policy numbers, email, and mobile number. Every figure in the report is an aggregate computed over de-identified rows. Biomarker and survey sample sizes (n values) appear throughout the report because the reader needs them to judge data quality; they are anonymous counts.

Clinical thresholds#

Biomarker prevalence uses fixed clinical cutoffs. Where WHI 2025 used a particular threshold, WHI 2026 matches it so the year-on-year comparison is methodologically clean. Where 2026 broadened a definition, the relevant page flags it.

The liver definition in 2026 counts any of SGPT >40, SGOT >40, or GGT >60 as liver dysfunction. This is broader than the single-marker cut WHI 2025 used, so the liver page reports its year-on-year comparison as directional rather than exact. The ESR cutoffs are approximate and not exactly comparable to 2025; the inflammation page is flagged as partial for that reason. The creatinine and uric-acid prevalence figures have no comparable 2025 published stat, so the kidney page presents them as 2026 baselines rather than trends.

Survey-side thresholds convert questionnaire answers into binary flags. Short sleep is the union of the "<5 hours" and "5–6 hours" answers. High stress is a self-reported 7 to 10 on a ten-point scale. Zero exercise is the "0" answer on weekly days of 30+ minutes of activity. Eating out three or more times a week is the union of the "3–4", "5–6", and "7+ times a week" answers. The WFO and sedentary-work flags map directly to the corresponding single answer option. Each lifestyle page states the exact value set it used.

Pincode-to-city mapping#

The biomarker file carries a clean six-digit pincode for every report. City is assigned deterministically by mapping the first three digits of the pincode to a reporting city, using the India Post pincode allocation scheme. The mapping is generous on metro boundaries: extended-metro pincodes fold into their core city. Thane, Navi Mumbai, Kalyan, and Vasai-Virar map to Mumbai; Noida, Ghaziabad, Gurgaon, Greater Noida, and Faridabad map to Delhi NCR; Pimpri-Chinchwad maps to Pune. The five WHI cities are Bengaluru, Delhi NCR, Mumbai, Pune, and Hyderabad; the table also covers Chennai, Kolkata, Ahmedabad, and roughly fifteen tier-2 metros.

Around 87% of the 11,427 biomarker reports map cleanly to a named city. Pincodes outside the city table fall back to a macro-region label derived from the first digit of the pincode (the postal zone), so no report is dropped. City-level biomarker findings are computed only on the mapped reports.

Year-on-year comparison#

WHI 2025 is the comparison baseline. Comparisons run against a hand-curated lookup of 2025 statistics — 63 entries covering the headline prevalence figures the 2025 report published — rather than against the raw 2025 microdata. Each 2026 page looks up the matching 2025 stat, computes the percentage-point delta, and reads the direction. Where the 2026 threshold differs from the 2025 threshold, the comparison is flagged as directional, not exact.

The WHI 2025 sample was smaller than 2026's across the board — roughly 3,437 survey respondents and 214,000 biomarker parameters, against 9,097 and 816,003 this year. Absolute year-on-year deltas should be read with that in mind. The effect is sharpest in city and gender sub-cuts, where the 2025 sample for a single city could number in the low hundreds.

Sample skews and limitations#

The datasets are useful and large, but each has structural limits. They are stated here in full so no page is read as claiming more than the data supports.

Every figure in this report traces back to a single reproducible analysis pipeline run over the three source datasets, so any number on any page can be regenerated and audited at source.