PBP inpatient psychiatric (cleaned view)

What it is

1. 1A semantic view over CMS Medicare Advantage Plan Benefit Package (PBP) filings exposing member cost-sharing for the inpatient psychiatric hospital benefit (PBP section B1B): per-plan, per-quarterly-snapshot copay structure (admission copay, per-day interval copays/day-counts, auth/referral requirements) for in-network inpatient psych stays.

Grain

How to read it

1. 1Compare what MA plans charge members for an inpatient psychiatric stay in the latest year.
2. 2Example: get the current admission copay per plan — SELECT contract_id, plan_id, segment_id, copay_amount, tier_1_day_interval_1_copay, tier_1_day_count_per_interval, auth_required FROM medicare_semantic.vw_pbp_inpatient_psychiatric WHERE report_year = 2026 AND quarter = (SELECT max(quarter) FROM medicare_semantic.vw_pbp_inpatient_psychiatric WHERE report_year = 2026).
3. 3To dedup a plan to its single most-recent snapshot across years, take the row with max (report_year, quarter) per (contract_id, plan_id, segment_id).
4. 4Use copay_amount/tier_1_day_admission_copay IS NOT NULL to restrict to plans with a captured numeric copay.

Traps

1. 1Snapshot stacking, no is_current flag: the same plan appears in multiple quarterly snapshots — 17,470 plan-years appear in ALL 4 quarters, and many span 2-3. A plan-level query that does not pin (report_year, quarter) will multi-count plans (e.g. 2023 has 31,005 rows for only 7,766 plans). For current state, pick the latest (report_year, quarter) per (contract_id, plan_id, segment_id) — that yields exactly 23,191 rows, one per plan-segment. report_year (NOT quarter) is the benefit-year axis; the profiler mislabeled 'quarter' as the time column.
2. 2cost_share_string is malformed: rows like '$900.0 copay + 900.0%' duplicate the copay dollar figure as a fake coinsurance percent (900% is impossible). The structured columns are correct (has_coinsurance=FALSE, coinsurance_pct=NULL there); never parse the '+ N.N%' suffix as real coinsurance. coinsurance_pct is 100% NULL table-wide, so no usable coinsurance value exists even when has_coinsurance=TRUE (flag-only).
3. 3Many governance/limit columns are dead in this view: coinsurance_pct, limit_text, limit_quantity, limit_unit, benefit_amount and all benefit_amount_* are 100% NULL; normalization_status/basis, aggregation_policy are 100% NULL; and network_type ('in_network'), max_oop_applies (true), offered_status ('offered'), numeric_amount_observability ('offered_without_numeric_amounts'), null_amount_reason, annualization_policy, combined_limit_policy, consumer_document_reconciliation_status ('not_checked'), confidence_tier ('source_faithful'), benefit_category, psych_tier_axis ('tier_1'), source_pbp_section/silver_table/lmi_authorities are all CONSTANT. Don't filter or pivot on these expecting variation.
4. 4numeric_amount_observability='offered_without_numeric_amounts' is misleading: it is constant and refers to the generic benefit_amount columns, yet ~56% of rows carry real dollars in copay_amount/tier_1_day_admission_copay. Use copay_amount/tier_1_* IS NOT NULL (109,095 rows) to find plans with a captured numeric copay — not this flag. Likewise has_copay=TRUE does not imply copay_amount is populated (26,615 such rows are NULL; 13,284 have has_copay itself NULL).
5. 5copay_amount is a duplicate of tier_1_day_admission_copay (equal in 100% of non-null cases) — they are the SAME number, not two independent cost components. The richer per-day structure lives in tier_1_day_count_per_interval + tier_1_day_interval_1_copay/interval_2_copay (e.g. ~$292/day for the first interval then $0). NULLs in auth_required (~29%), referral_required, cost_on_discharge_day (~56%), and hospital_benefit_period_code (~37%) mean 'not captured / unknown', NOT false — and NULL hospital_benefit_period_code correlates with the no-numeric-copay rows. version is pre-deduped (one per grain key); ignore it as a selector. This view is the in-network, tier-1 psych slice only — no out-of-network and no higher psych tiers.

Columns (173)

Query it

describe_data_source(source_fqn="medicare_semantic.vw_pbp_inpatient_psychiatric")     # the card, live
query_substrate(sql="SELECT * FROM medicare_semantic.vw_pbp_inpatient_psychiatric WHERE contract_id = '<value>' AND plan_id = '<value>' AND segment_id = '<value>' AND report_year = '<value>' AND quarter = '<value>' LIMIT 100")