48 verified sources.
59 PDFs archived.

Authoritative source for the 2025 SF death toll: 625 accidental overdose deaths — the finalized OCME count, not the preliminary 621 cited in early news reports. Documents complicating substances: Fluoro Fentanyl (38), Xylazine (27), Bromazolam (11). Lowest annual toll since systematic OCME tracking began.

2023: 810 deaths — confirmed all-time record high. 2024: 635 deaths. Together with 2020–2022 reports, these establish the complete six-year trajectory. 2023 report confirms 81% fentanyl involvement (653 of 810 deaths per UCSF analysis).

Complete trajectory baseline: 726 (2020) → 642 (2021) → 649 (2022). Establishes the pre-peak trend and confirms consistent geographic concentration in Tenderloin/SoMa. Used for ACS trend calibration and crisis framing throughout the platform.

Primary source for the 5× Black/African American overdose disparity — confirmed verbatim: "an opioid overdose death rate that is more than five times higher than the citywide rate." Strategic Area 4 lists Medical Examiner report review as the primary data mechanism — confirming SF operates on lagged, reactive surveillance with no real-time forecasting component.

Independent corroboration of racial disparity: "Black/African Americans in San Francisco experience fatal overdoses at 5-times the city-wide rate." Second official source confirming this figure, separate from the ODP 2024.

Longform investigation confirming: "by the end of 2023, the annual overdose death toll was 810, the highest ever recorded." Documents that fentanyl killed more than 3,000 SF residents in its first decade in the city. Key source for peak-year framing.

Documents enforcement-driven displacement from Sixth Street to the Mission district. Reporters observed ~50 people using and selling drugs at 16th Street BART following Sixth Street clearance. Primary source for geographic displacement patterns — the key driver of the STI and ACS displacement-tracking use case.

Documents 40% year-over-year increase in overdose calls to Nob Hill through September 2023. Geographic spread to Western Addition, Bayview-Hunters Point, and Mission documented with SFFD dispatch data. Empirical basis for multi-neighborhood ACS coverage requirement.

UCSF ethnography documenting transition from injection to smoking; fentanyl priced at $10/gram. Identifies fentanyl residue in shared smoking equipment as a novel overdose vector. Confirms 653 fentanyl-related deaths in 2023. Supports STI design rationale for smoking-route emergence detection.

Cross-domain source (D1 and D2). Of 617 re-analyzed 2022 cases, 54 (9%) had Novel Synthetic Opioid and/or Xylazine newly identified. Confirms Fluoro Fentanyl potency "ranges from half to up to five times the potency of fentanyl." Validates the need for comprehensive toxicological re-analysis beyond standard panels.

Primary pharmacological source for cychlorphine. Confirmed verbatim from PDF: "In vitro pharmacology data show this drug to be approximately 10x more potent than fentanyl [Vandeputte & Stove, personal communication]." Potency is from personal communication — not peer-reviewed clinical pharmacology. Documents 25 fatal overdose specimens in the US as of January 2026, including cases from California.

Confirms first SF cychlorphine death April 2026. Dr. Philip Coffin (DPH): cychlorphine "is not detected on the available fentanyl test strips." Examiner states naloxone "is believed to help reverse the effects" — note this conflicts with ABC7 (D2-03) which says "resistant to Narcan." MERIDIAN uses: multiple doses may be required, partial effectiveness believed.

DEA Special Agent Bob Beris (SF Field Division) confirmed cychlorphine in Northern California counterfeit pills. Dr. Coffin confirmed test strip limitation verbatim. DPH Director Tsai: "We believe it is more potent than fentanyl." ABC7 states it "is believed to be resistant to Narcan" — inconsistency with Examiner source documented and reflected in STI guidance.

DEA confirmed cychlorphine in two Northern California samples. First federal identification: April 2024 in Florida. 35 samples nationwide as of April 2026. Multiple prior deaths in Tennessee. Documents the spread trajectory that makes cychlorphine an active SF threat, not a theoretical one.

Official confirmation of xylazine in SF. Detection window confirmed verbatim: "mid-December 2022 and mid-January 2023." Key clinical statement confirmed: "Naloxone will reverse the opioid but does not reverse xylazine effects." All four decedents also had fentanyl detected.

Formal California EMS alert confirming nitazenes are "considerably more powerful than fentanyl and may require significantly higher doses of naloxone to reverse." Issued by the San Diego County EMS Medical Director. Key basis for STI's nitazene tracking parameters and naloxone dosing guidance.

ODMAP is free, web-based near-real-time overdose surveillance. Confirmed from PDF: "over 5,300 agencies in all 50 states... as of April 2025, nearly 3 million overdose events have been entered." Spike alerts fire when a designated area exceeds a threshold in a rolling 24-hour window. SF has no equivalent.

NIJ documentation of ODMAP's real-time capability. Data entry "takes seconds." In Berkeley County, WV, ODMAP showed nearly 20% of overdoses at a single location, enabling targeted response. ODMAP spike alerts show Baltimore City spikes are followed 8–12 hours later by spikes in neighboring jurisdictions — demonstrating cross-jurisdiction intelligence value SF lacks.

The primary published comparator for EMERGENZ Meridian. ML-based forecasting at census block group level in Rhode Island. 39 municipalities randomized. PROVIDENT identifies highest-risk 20% of CBGs (162 of 809) for targeted resource deployment. MERIDIAN's ACS and STI are structurally modeled on PROVIDENT's validated architecture.

Documents Rhode Island's complete overdose data dashboard ecosystem including PROVIDENT, community co-design methodology, and real-time surveillance integration. Emphasizes "real-time surveillance and rapid data-sharing needs." Community organizations directed resources to high-risk neighborhoods identified by forecasting model. Demonstrates that data dashboards require ongoing community partnership — a principle built into MERIDIAN's credentialed access model.

Validated county-level predictive modelling. Mixed-effects negative binomial regression correctly identified 41.6–56.8% of top-decile mortality counties. Key predictors: geospatial proximity, opioid prescription rates, healthcare access, socioeconomic factors. San Diego State/UC San Diego/Oxford/UCSF. Foundational evidence for MERIDIAN's prospective predictive architecture.

Demonstrates the limitations of relying solely on ME data for SF overdose surveillance. Reviews 816 opioid overdose deaths (2006–2012). Documents the granularity and lag inherent in ME-dependent surveillance. Primary evidence that reactive ME data is insufficient for real-time public health intelligence. Note: Elise D. Riley is not an author — confirmed from PDF.

Strategic Area 4 of the plan — "Track overdose trends and related drug use metrics to measure success and inform program development" — explicitly relies on Medical Examiner report review as its primary data mechanism. No mention of real-time surveillance, predictive modeling, or near-real-time alert infrastructure. This documented absence is the core justification for MERIDIAN.

Demonstrates the level of EMS transparency available in peer jurisdictions. King County EMTs responded to 250,000+ calls in 2024; paramedics responded to 48,000+ ALS calls. Publishes response time, cardiac arrest survival, and overdose intervention data by neighborhood annually. Contrasted with SF's absence of comparable public EMS intelligence — evidence for the EES infrastructure gap.

Foundational source for ACS 311 composite weights. Columbus, OH 2008–2017. Confirmed verbatim from PDF: "code violation, public health, and street lighting were the top three accurate predictors with predictive accuracy as 0.92, 0.89 and 0.83, respectively." 10 of 21 311 request types spatially associate with overdose events. Note: v1.0 weights are Columbus-derived; SF-specific validation is planned for v1.1.

Spatial clustering analysis of heroin overdose incidents in Cincinnati at census block group level using Ripley's K function and Moran's I. Identified 7 hot spot clusters. High-overdose neighborhoods: higher crime, higher poverty, lower education, lower income, greater distance to treatment. Validates spatial clustering methodology and sociodemographic correlates informing ACS component selection.

NYC 311 data as NLP blight detection algorithm input. Algorithm sensitivity ~90%. Blight-related 311 calls most strongly correlated with long-term commercial vacancies. Validates the methodological framework of using 311 civic complaint data as a public health signal — the conceptual basis for the ACS 311 composite component. (Note: originally filed as "file.pdf" in the evidence base.)

Technical documentation for the nationally validated CDC/ATSDR SVI — 16 US Census variables across 4 themes. Used as the SVI component in the ACS (weight 0.15). Confirmed download URL from PDF: https://www.atsdr.cdc.gov/placeandhealth/svi/data_documentation_download.html (note: differs from some third-party references; use this confirmed URL).

Compares 13 composite neighborhood indices across 57,000+ US census tracts. Key finding: "indices are not interchangeable" — supporting the ACS multi-component approach over reliance on any single index. Composite indices are gaining prominence in equity research and health policy. Direct methodological justification for MERIDIAN's multi-layer ACS architecture.

Technical reference for the Getis-Ord Gi* spatial statistic used in hot spot analysis — the methodological foundation for spatial clustering in both Li et al. 2020 (Columbus) and Choi et al. 2022 (Cincinnati). Documents that statistically significant hot spots are features with high values surrounded by other high values. Used to document the established spatial analysis methodology that MERIDIAN's geographic scoring follows.

The actual SF 311 data source for the ACS 311 composite component. Georeferenced service requests with category types, timestamps, and location data. Updated nightly. Public domain license. Available via DataSF API for automated ingestion. Confirms the 311-based ACS component is operationally feasible in San Francisco — the Columbus methodology applies directly to available local data.

Primary source for the SRO housing component in the ACS. Confirmed verbatim from PDF highlights box: "Overdose mortality among SRO residents was 19 times higher than among non-SRO residents." Exact figure: "19.3 (95%CI 17.1–21.7) times." SRO residents comprised ~3% of SF adult population but experienced dramatically elevated mortality. Study period 2010–2017.

LASSO and random forest algorithms applied to 203 ACS covariates for 742 Rhode Island census block groups (2016–2019). Identified neighborhood-level predictors: education attainment, income, residential stability, racial/ethnic composition, social isolation. Validates the use of ACS variables — the same data used in MERIDIAN's ACS — as legitimate predictors of overdose mortality at census block group level.

Network analysis of ~500,000 opioid overdose events across 481 counties (2018–2023). 6.3% of events occurred outside person's county of residence. ~10% of discordant events over 204 km from home. Demonstrates resource deployment based on residential population data is misaligned with actual overdose location distribution — evidence for the ACS's displacement-tracking design.

Spatial social network analysis of Milwaukee, WI overdose deaths (2017–2020). 26.72% of deaths geographically discordant. "Hub" communities predominantly White; "authority" communities lower housing stability, higher poverty. Temporal trend analysis identified consistent, sporadic, and emergent hotspots. Authors: Forati, Rina Ghose, Mohebbi, Mantsch (confirmed from PDF).

Validated GIS-based composite scoring tool for all California census tracts. 21 indicators across pollution burden (11) and population characteristics (10) domains. Used as the CalEnviroScreen component in the ACS (weight 0.05) — capturing environmental pollution exposure as a long-term stressor associated with substance use vulnerability. Statewide validated methodology directly applicable to SF census tracts.

Systematic review of 28 studies (pre-July 2022). Findings: socioeconomic conditions and drug overdose death rates are moderately but consistently associated across a large number of studies and measures. Employment, income, and poverty interventions identified as effective targets. Strongest available synthesis evidence that area-level deprivation indices — the core of the ACS — have a validated evidence base for predicting overdose mortality geography.

Duke University Hospital EHR retrospective cohort. ADI associated with drug-related admissions but did not fully explain spatial variation — demonstrating that single-variable deprivation indices are insufficient. Key evidence for why a composite approach outperforms reliance on any single socioeconomic index.

Scoping review of 15 commonly used area-level deprivation indices in US public health research since 2015. Key finding: composite indices have stronger relationships with health outcomes than single-variable measures. Composite indices are gaining prominence in equity research and health policy decision-making. Direct methodological justification for MERIDIAN's multi-component ACS architecture.

PROVIDENT's use of census block group-level ACS variables as predictors of overdose mortality directly validates MERIDIAN's ACS approach. PROVIDENT's 809 RI CBGs / 162 highest-risk identified mirrors MERIDIAN's method of scoring all SF CBGs and identifying elevated-risk areas for resource targeting.

Massachusetts population-based retrospective cohort. Confirmed verbatim from PDF: "Of the 11,557 patients who met study criteria, 635 (5.5%) died within 1 year." Basis for EES post-overdose re-contact rate parameter. A prior version of the MERIDIAN methodology cited "7–18%" — this had no traceable source and has been corrected to the verified 5.5%.

Primary source for EES Unit Hour Utilization benchmark. Confirmed verbatim from ICMA PDF: "EMS agencies responding solely to 911 calls typically target a lower unit hour utilization (between 0.30 and 0.50 UHU)." The 0.30–0.50 range is the EES reference benchmark for interpreting UHU relative to overdose call density.

Documents Project FRIEND — SF Fire Department EMS leave-behind naloxone program. Sept 2019–Sept 2020. 1,200 kits distributed; 232 (19%) registered; 146 (63% of registered) distributed during suspected overdose encounters. Demonstrates SF EMS system's existing naloxone deployment infrastructure — the operational basis for the NCS leave-behind deployment layer.

Authoritative source for SF EMS response time standard. Confirmed verbatim from PDF: "ambulances should arrive at the scene of a life-threatening emergency medical incident within ten minutes at least 90 percent of the time." A prior MERIDIAN document cited an "EMDAC 4-minute ALS standard" — this was unverifiable and has been corrected to this confirmed SF EMSA standard. SFFD has met the 90% target in five of the last seven fiscal years.

Documents naloxone vending machine locations at VA SF facilities — the only publicly confirmed accessible naloxone vending machines in SF as of May 2026. Confirmed locations: SF VA Medical Center (4150 Clement St — 2 machines), SF VA Downtown Clinic, Oakland VA Clinic, Oakland VA Behavioral Health Clinic, South Santa Rosa Clinic. Machines dispense free naloxone kits. Absence of comparable public-facing vending for the general population is a documented NCS gap.

SF-specific naloxone geography study. 195 SF census tracts (2010–2012). Greater distance to nearest naloxone distribution site associated with lower naloxone reversals [IRR = 0.51 per 500m increase, 95% CI 0.39–0.67, p<0.001]. Direct evidence base for the NCS 400-meter coverage radius and geographic naloxone gap analysis. UCSF/SFDPH collaboration.

Baltimore, MD spatial analysis of 518 overdose events. Naloxone administration inversely associated with distance to nearest distribution site [IRR = 0.72 per 1,000m, 95% CI 0.59–0.89, p = 0.002]. Confirmatory evidence from a second major US city alongside Rowe et al. 2016 (SF) that proximity to naloxone distribution predicts utilization. Johns Hopkins.

SF DPH 2024 plan documents: naloxone distribution increased from 47,000 doses to 158,000+ doses in FY 23–24. 4,126 people trained in overdose recognition and naloxone use. Despite this volume, the absence of geographic distribution mapping relative to overdose demand is the NCS's core problem statement: volume does not equal coverage.