Preventable Hospitalizations (OASIS Ambulatory Care Sensitive Conditions)

Limitations

OASIS's percent of all inpatient stays that are for a preventable (ambulatory care sensitive) condition, not a rate per resident. Stays of Georgia residents in Georgia's non-federal hospitals only. Conditions nest: never add groups and conditions together. The October 2015 switch to ICD-10-CM breaks several conditions' series; three have no rows after it. Race groups start in 2016 (NULL above 20% unknown). OASIS withholds the percent on 1-4 stays (NULL). The conditions are OASIS's own list (its ACSC definitions page): conditions that timely, effective outpatient care can often keep out of the hospital, used as prevention quality indicators. It is not AHRQ's Prevention Quality Indicators (our reading: it includes dental conditions, failure to thrive and iron deficiency anemia, sets age limits only on a few conditions, and gives a percent, not a rate per resident). The percent divides by all of the group's stays (every cause, childbirth and newborns included), so it shows how much of a place's hospital use is for these conditions: a place with many births or surgeries shows a lower percent for the same preventable stays. A stay counts in one condition at most (OASIS reports the first one found). A discharge is one inpatient stay in a non-federal acute-care hospital in Georgia, counted in the year the patient left (alive or dead), by the patient's county of residence: a person admitted twice counts twice, newborns' birth stays and deliveries are included, and stays of Georgia residents in other states' hospitals or in federal (VA and military) hospitals are not, so counties near another state's hospitals may look low. 41% of rows with at least one preventable stay have no percent (1-4 stays, mostly small counties, groups and conditions). Where a group had no stays at all in a place (num_discharges_all_causes 0) the percent is NULL: OASIS shows 0, but a percent of nothing is undefined (our mask, 141,264 rows). ICD-9-CM codes until September 2015 and ICD-10-CM from October 2015, 2015 mixing both: Georgia's bacterial_pneumonia 21,426 in 2014 and 59,398 in 2017; iron_deficiency_anemia 16,833 in 2014 and 191 in 2017; kidney_or_urinary_infection 8,393 in 2014 and 2,128 in 2017; hypertension 2,528 in 2014 and 443 in 2017; diabetes_without_complications 1,393 in 2014 and 58 in 2017; diabetes_with_other_complications 4,289 in 2014 and 7,461 in 2017; congestive_heart_failure 28,266 in 2014 and 16,759 in 2017; vaccine_preventable_conditions 240 in 2014 and 46 in 2017. The groups carry their conditions' breaks: avoidable_conditions 28,644 and 20,074, acute_conditions 82,685 and 125,332, chronic_conditions 72,906 and 58,030, all conditions 184,235 and 203,436. Three conditions fall to 0 in every place, group and payer, and no stay can carry them from then, so they have no rows (dropped, not served as zeros): cervical_cancer (160 stays in 2015; no rows from 2016); congenital_syphilis (15 stays in 2015; no rows from 2016); pulmonary_tuberculosis (63 stays in 2016; no rows from 2017). OASIS says cervical cancer is no longer an ACSC in ICD-10; the other two have ICD-10-CM codes, but no stay carries OASIS's definitions (our reading). failure_to_thrive falls from 146 stays in 2015 to 0 from 2016 except 1 in 2021, so it keeps its rows; its 0s are true counts under OASIS's definition, which rarely captures failure to thrive in ICD-10-CM codes (our reading): they do not mean the condition went away. skin_grafts_with_cellulitis (defined by DRG) is 0 in every place in 2013 and 2019 (159, 0, 168 stays in 2012-2014; 155, 0, 190 in 2018-2020): not real. nutritional_deficiencies, which count secondary diagnoses too, grew from 1,404 stays (2000) to 30,423 (2024) and 39,775 (2025), with a jump in 2008 (2,064 to 4,950), and drive avoidable_conditions (12,403 stays in 2000, 40,946 in 2025): probably more coding of malnutrition, not more of it (our reading). convulsions and epilepsy trade stays in 2007-2008 (convulsions 4,292 in 2006, 2,417 in 2008; epilepsy 1,705 and 3,777), a coding move (our reading). In 2025, the newest year, Georgia's preventable stays rose to 226,635 (19.82% of all stays) from 205,497 (18.58%) in 2024, led by nutritional deficiencies (30,423 to 39,775) and dehydration (46,309 to 52,785); OASIS documents no reason. OASIS's hospital data records no ethnicity before 2016; from 2016, 4.7% to 15.6% of Georgia's preventable stays a year are race_unknown (mostly unknown ethnicity), up to 98% of a county's in a year, and stays of unknown ethnicity are more often preventable (28.8% in 2024, against 18.6% for all), so the race groups' percents may run low (our reading). race_unknown, all minus the seven race groups (our calculation, both counts), is in every row OASIS's own unknown-ethnicity group, since OASIS records no non-Hispanic stay of unknown race; its percent is OASIS's. Where race or ethnicity is unknown for more than 20% of a place-year's stays (race_unknown / all, all causes), the race groups' counts and percents for that place and year are NULL (our mask, OASIS's own 20% threshold for missing items): each group's counts would miss its share of the unknowns, so its figures follow the recording, not the group. The all, sex, payer and race_unknown rows are kept, so the unknown share stays visible. It applies to 168 county-years (61 counties, never Georgia), the same as hospital_discharges_by_cause's. Race groups follow the platform convention from OASIS's race x ethnicity cross: hispanic is Hispanic of any race; white, black, asian, native_american (American Indian or Alaska Native), pacific_islander (Native Hawaiian or Other Pacific Islander) and multiracial are non-Hispanic. A payer's percent divides by that payer's own stays. The payer is the primary payer as recorded on each record, and coding shifts show as jumps between payers. OASIS counts HMO and other managed care under private insurance and does not say where Medicare Advantage or Medicaid managed-care plans fall (our caution). PeachCare has no rows before 2006 (no stay recorded it, so no value can exist). OASIS's payer and race breakdowns are never crossed. Not served: OASIS's SES-vulnerability rows (their percents are valid, each dividing by the SES group's own stays or visits, but the platform has no SES group yet), age and education breakdowns. OASIS's reporting-issues page notes: no 2012 data from Hart County Hospital (Hart and nearby counties low); Candler County's hospital did not report in 2013 (Candler low; Bulloch, Evans, Emanuel and Tattnall slightly); and the ICD-10-CM switch (October 2015), after which some conditions 'appear to be out of trend' (https://oasis.state.ga.us/dtt/content/KDIACSC.html). The conditions are defined differently from hospital_discharges_by_cause's causes (asthma there is not the same count as asthma here). 2025 was added to OASIS on 2026-09-03; OASIS marks no year provisional. Counties add up to Georgia, except race groups from 2016, whose counts are NULL in the county-years the race mask covers (Georgia's race rows count every county, so a race group's counties add up to less there). tract_geoid is always NULL.