Internal protocol
Frozen
Frozen 8 Sept 2026, 4:29 pm
Prospective study protocol · v1.1.1Primary collection period
Persistent site-level contrasts in hourly PM2.5 across Delhi-NCR relative to the concurrent network median: a prospective observational study
The protocol sets out the primary 2026–27 analysis before that study period begins. Data collected during September are retained for method development but are not part of the primary analysis.
v1.1.1 is the current external-registration copy. The full revision history is recorded below.
External registry
Registered under embargo
OSF Registries
Primary window
1 Oct 2026
through 31 Mar 2027 · Delhi local time
Baseline sites
73
Frozen 1 Oct 2026, 12:30 am
Primary research question
How large are persistent site-level departures from the concurrent Delhi-NCR network median of hourly ambient PM2.5 during the prespecified primary period?
Primary outcome: Hourly PM2.5 concentration (µg/m³) at the physical monitoring site × Delhi-local hour level. The concurrent network median is used as a same-hour statistical reference; it is not an official city average, regional-background estimate or personal-exposure measure.
Study periods
Pilot and primary data are analysed separately.
Pilot
6 Sept 2026 → 30 Sept 2026
Used for source checks, QC development and exploratory review. These observations are excluded from the primary analysis.
Primary
1 Oct 2026 → 31 Mar 2027
Prespecified analysis window. It ends on 31 Mar 2027 regardless of the results observed; routine collection may continue afterward.
External registration
OSF registration linked.
The internal protocol is frozen and versioned. External OSF registration provides an independent timestamp and should be completed before the primary study period begins.
Protocol history
Version changes are preserved rather than overwritten.
All revisions shown below occurred before the primary study period. v1.1.0 narrowed the framing after a pre-registration literature review and added two descriptive monthly persistence analyses; v1.1.1 only corrected a stale internal version label. The primary estimand, dates, thresholds and core analysis rules were unchanged.
v1.0.0
Initial frozen protocol
8 Sep 2026
Established the prospective study window, primary residual estimand, completeness rules, QC, missingness, sensitivity analyses and bootstrap procedure.
v1.0.1
Editorial revision
8 Sep 2026
Rewrote the protocol in plainer academic language. The scientific design and analysis rules were unchanged.
v1.1.0
Literature-positioning amendment
8 Sep 2026
Narrowed the framing to network-relative site contrasts and added prespecified descriptive month-to-month persistence summaries. The primary estimand and core rules were unchanged.
v1.1.1Current
Current registration copy
8 Sep 2026
Corrected one stale internal version label. Scientifically identical to v1.1.0 and used as the current external-registration copy.
Study positioning
What this study is designed to add
Prior Delhi and Delhi-NCR studies have already documented spatial heterogeneity, long-term temporal trends, meteorological associations, and fine-scale pollution variation. This study does not claim that spatial variation itself is new.
The primary analysis estimates each physical monitoring site's repeated same-hour departure from the concurrent Delhi-NCR network median, using a frozen physical-site cohort and explicit shared-hour completeness rules. The network median is used as a contemporaneous reference, not as an official city average or an estimate of regional background.
Secondary persistence analysis
Month-to-month stability
For each physical site and Delhi-local calendar month, calculate the median R_it within that month when the site has valid PM2.5 for at least 70% of the included primary network hours in that month.
For sites with at least 3 eligible calendar months, report the proportion of eligible months in which the monthly median residual has the same sign as the full-period primary site effect. A zero monthly median is counted as neither positive nor negative and therefore as not sign-concordant.
For each pair of calendar months with at least 10 shared eligible physical sites, report Spearman rank correlation between monthly site effects. These correlations are descriptive; no binary significance threshold is used as a study conclusion.
Primary estimand
Primary spatial comparison
Rit = PM2.5it − medianj(PM2.5jt)
Site-hour residual
R_it = PM2.5_it - median_j(PM2.5_jt), where j denotes eligible physical monitoring sites reporting at hour t.
Site effect
For each eligible physical site i, the primary site-level effect is the median R_it across all included primary-period hours for that site.
Network heterogeneity
Spatial heterogeneity across the monitoring network is summarized by the interquartile range and the 90th–10th percentile range of the site-level effects.
Uncertainty
After at least 30 represented primary-period days are available, 95% percentile confidence intervals will be calculated by block-bootstrap resampling Delhi-local calendar days 1,000 times using deterministic seed 20261001.
Eligibility
Eligibility rules
Baseline cohort
At the start of the primary period, baseline eligible physical sites are those marked active in the station registry with at least one valid PM2.5 observation during the preceding 7 Delhi-local calendar days. This baseline cohort is then frozen for the primary analysis.
Network hour
A network hour is included when at least 50% of baseline eligible physical sites have a valid PM2.5 value in that hour.
Site inclusion
A physical site is included in the primary site-level summary when it has valid PM2.5 values for at least 70% of included primary network hours.
Missingness
Missing observations remain missing. No interpolation, carry-forward, or substitution from a neighboring monitor is used.
Quality control
Prespecified quality-control rules
Critical flags
An observation with an observation-specific critical QC flag is excluded from the primary analysis. The underlying raw and provenance records remain retained.
Review flags
Review-flagged observations remain in the primary analysis unless the flag identifies the normalized value as invalid. A prespecified sensitivity analysis excludes all review-flagged observations.
Outliers
No discretionary concentration outlier deletion is allowed. High source values remain unless they meet a prespecified QC exclusion rule.
Station identity
Station renames, operator changes, and coordinate shifts are retained as audit events. If a metadata change suggests that observations before and after the change may not represent the same physical site, a protocol amendment is required before they are combined in the primary analysis.
Secondary analyses
Prespecified descriptive extensions
- 1PM10 analyzed using the same descriptive framework as a secondary pollutant outcome.
- 2Delhi-local diurnal PM2.5 summaries based on network-hour medians grouped by local clock hour.
- 3Station-to-station Pearson correlations using shared hourly PM2.5 observations, with same-physical-site pairs excluded from independence-sensitive summaries.
- 4Paired campus-reference monitor differences using only hours observed at both monitors. Campus comparisons are a secondary application and are not the primary scientific estimand.
- 5Descriptive comparison of Delhi-only and wider Delhi-NCR results.
- 6Monthly site-level median residuals using the primary R_it definition, requiring at least 70% site completeness within included network hours for that Delhi-local calendar month.
- 7Temporal persistence of network-relative site contrasts: for sites with at least 3 eligible months, report sign concordance of monthly versus full-period site effects; for month pairs with at least 10 shared eligible sites, report Spearman rank correlation of monthly site-effect rankings. These are descriptive secondary measures.
Sensitivity analyses
Prespecified alternate analysis rules
- 1Exclude all review-flagged observations.
- 2Repeat site inclusion using 60% and 80% completeness thresholds instead of 70%.
- 3Restrict the analysis to Delhi only.
- 4Use one deterministic source feed per physical site instead of the within-site hourly median.
- 5Use mean site-hour residuals instead of median site-hour residuals.
Weather-adjusted extension
Meteorological adjustment remains gated
No weather-adjusted model will be fitted until at least 30 represented air-data days and 21 days of protocol-eligible locked-model meteorology are available.
The exact model formula, transformations, diagnostics, and inferential procedure must be specified in a dated protocol amendment before the first weather-adjusted model is fitted. Weather-adjusted analyses are not part of the primary estimand in this protocol.
Interpretation boundary
Statistical interpretation
The primary analysis is descriptive and interval-based. No single binary null-hypothesis test is designated as the primary conclusion.
Differences between monitoring sites are not interpreted as evidence of a causal local emission source, personal exposure, a campus-specific exposure effect, or a health effect.
Ethics & scope
Environmental data, not patient data.
The study uses publicly available area-level environmental measurements and contains no individual-level patient or participant data. No formal ethics exemption is claimed in advance. Any determination required by an institution or journal will be sought. Any future linkage to individual-level or non-public health data would require separate ethics, privacy, and data-governance review.
Protocol identity
DAH-PM25-2026-27-v1-1-1
SHA-256 22076bd8e22dba94f8f6ca11bb66f0ac09fe4e9dc6801a1a6ebd76682835672a