To understand pain medication patterns around the time of delivery, we studied more than 7 million deliveries that occurred between 2017 and 2024. We analyzed how discharge prescriptions for opioids have changed over time. We found that there has been a substantial decline during the studied period in the proportion of women who received an opioid prescription upon discharge, as seen in Figure 1. In 2017, more than half of all patients were prescribed an opioid upon discharge. By 2024, that figure had dropped to roughly a quarter of deliveries, representing a 43% decrease.

We also analyzed how the use of pain medications varies throughout the delivery admission. We categorized the stages as pre-labor (in hospital), labor and delivery, and post-delivery (in hospital). Opioid medications included those administered orally, intravenously, intramuscularly, subcutaneously, sublingually, transdermally, or epidurally. Data on patient-controlled analgesics (PCAs), which are used during labor and delivery in some situations,¹ were not available for analysis.

Opioid use varied by whether the mother had an opioid prescription in the 30 days prior. 76% of women with a recent opioid prescription received an opioid during labor compared to 65% of those without such a history, as seen in Figure 2. This trend was similar for before and after labor as well.

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson.

1. Reale S. Pharmacologic management of pain during labor and delivery. UpToDate. February 26, 2024. https://www.uptodate.com/contents/pharmacologic-management-of-pain-during-labor-and-delivery. Accessed May 5, 2025.
2. Committee Opinion No. 711: Opioid Use and Opioid Use Disorder in Pregnancy. Obstet Gynecol. 2017;130(2):e81-e94. doi:10.1097/AOG.0000000000002235

To further understand RSV vaccine effectiveness in older adults, we studied 1,204,649 patients aged 60 and older who had an RSV lab test during the 2024/2025 RSV season. Patients were matched based on the month of their RSV lab test, age, and factors that indicate increased risk of complications from an RSV infection. Patients were grouped based on the time elapsed between RSV vaccination and RSV lab testing.

During the 2024/2025 RSV season, patients vaccinated two weeks to six months prior had a positive RSV infection rate of 1.1%, while those vaccinated more than a year prior had rates more than twice as high and those who have never received the RSV vaccine had rates nearly four times as high. Similar patterns were observed for ED visits and hospitalizations, as seen in Figure 1.

Vaccine effectiveness (VE) declined steadily over time. For RSV infection, VE was 71% at 4 months, decreasing to 40% by month 19. Effectiveness against RSV-related ED visits followed a similar pattern, dropping from 71% in month 4 to 46% by month 19. Hospitalization protection declined from 73% to 44% over the same period.

These findings suggest that while protection against severe outcomes persists into the next season, it is significantly reduced compared to the initial vaccination season.

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson.

1. Bartelt K, Deckert J, Gracianette M, Barkley E. RSV Vaccine Can Prevent More Than 70% of RSV Infections, ED Visits, and Admissions Among Older Adults. Epic Research. https://epicresearch.org/articles/rsv-vaccine-can-prevent-more-than-70-of-rsv-infections-ed-visits-and-admissions-among-older-adults. Accessed on June 30, 2025.

To further explore the relationship between SSRI use and cancer diagnoses across various cancer types, we studied 627,964 adult patients, including patients with no history of SSRI use and those with three to eight years of SSRI use. Patients were matched 1:1 based on legal sex and age, and we accounted for other factors such as demographics, smoking history, BMI, use of other antidepressants, and comorbidities in our analysis.

We found that SSRI use did not significantly change the overall risk of developing cancer, as seen in Figure 1. Similarly, no significant associations were observed for lung, breast, colorectal, prostate, pancreatic, or melanoma cancers. However, patients on SSRIs had an 18% lower risk of hematologic cancers compared to those not on an SSRI. This aligns with prior animal studies suggesting SSRIs may enhance immune surveillance by modulating T-cell activity.¹

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson.

1. Li B, Elsten-Brown J, Li M, et al. Serotonin transporter inhibits antitumor immunity through regulating the intratumoral serotonin axis. Cell. Published online 2025. doi:10.1016/j.cell.2025.04.032
2. Boursi B, Lurie I, Mamtani R, Haynes K, Yang YX. Anti-depressant therapy and cancer risk: a nested case-control study. Eur Neuropsychopharmacol. 2015;25(8):1147-1157. doi:10.1016/j.euroneuro.2015.04.010
3. Ashbury JE, Lévesque LE, Beck PA, Aronson KJ. A population-based case-control study of Selective Serotonin Reuptake Inhibitors (SSRIs) and breast cancer: the impact of duration of use, cumulative dose and latency. BMC Med. 2010;8:90. Published 2010 Dec 22. doi:10.1186/1741-7015-8-90

To understand trends in ASD diagnoses over time, we studied 338,415 patients with a pattern of established care who received their first ASD diagnosis between 2015 and 2024. We found that the overall average age at ASD diagnosis remained stable at around 10.5 years, while the median age decreased from 7 in 2015 to 6 in 2024. When the average is higher than the median, it typically reflects a small but significant group being diagnosed much later, potentially into adulthood, which pulls the average upward.

Among male patients, diagnoses are increasingly occurring earlier, with the median age at diagnosis dropping from 7 to 5 and the average age dropping from 10.2 to 9.2 over the study period. In contrast, the median age for female patients has stayed consistent at about 8 years, while the average age at diagnosis increased from 11.7 years in 2015 to 13.1 years in 2024.

We then examined the age distribution at the time of autism diagnosis. Early detection has become more common among male patients, with 44% diagnosed before age 5 in 2024, compared to 34% of females. In contrast, nearly one in four female patients with ASD were diagnosed as adults (age 19 or older), more than double adult diagnoses among males (12%). More than half of both males and females were diagnosed after age 5 and might have benefitted from earlier diagnosis.

These findings underscore ongoing improvements in early childhood autism detection, particularly for boys, while highlighting a parallel rise in adult diagnoses among women. Persistent differences in age at diagnosis by sex suggest that symptom presentation and screening continue to differ in males and females.

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson.

1. Lipkin PH, Macias MM; COUNCIL ON CHILDREN WITH DISABILITIES, SECTION ON DEVELOPMENTAL AND BEHAVIORAL PEDIATRICS. Promoting Optimal Development: Identifying Infants and Young Children With Developmental Disorders Through Developmental Surveillance and Screening. Pediatrics. 2020;145(1):e20193449. doi:10.1542/peds.2019-3449
2. Hyman SL, Levy SE, Myers SM; COUNCIL ON CHILDREN WITH DISABILITIES, SECTION ON DEVELOPMENTAL AND BEHAVIORAL PEDIATRICS. Identification, Evaluation, and Management of Children With Autism Spectrum Disorder. Pediatrics. 2020;145(1):e20193447. doi:10.1542/peds.2019-3447
3. McCarty P, Frye RE. Early Detection and Diagnosis of Autism Spectrum Disorder: Why Is It So Difficult?. Semin Pediatr Neurol. 2020;35:100831. doi:10.1016/j.spen.2020.100831

To examine whether blood glucose monitoring influences average blood glucose as measured by HbA1c, we studied 9,675 adults without evidence of diabetes who had a baseline HbA1c between 5.0% and 5.9%. Lower HbA1c values represent better blood glucose control over time, while higher values represent less controlled blood glucose levels. Patients were categorized into three groups: those prescribed CGMs, those prescribed intermittent glucose monitoring (such as testing strips), and those who received neither. All patients had a follow-up HbA1c result between 6 and 18 months later. Patients were matched by year and time between their HbA1c labs in the study period.

We found that average HbA1c 6 to 18 months following an initial HbA1c showed negligible changes regardless of glucose monitoring method. Patients prescribed a CGM had their average HbA1c rise from 5.49% to 5.50%, as seen in Figure 1. Patients prescribed intermittent glucose monitoring had a slightly greater increase, from 5.55% to 5.62%. Among patients with no glucose monitoring, a slight rise was also observed, from 5.55% to 5.57%.  

A sensitivity analysis accounting for factors such as demographics and baseline HbA1c found similar results. 

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson.

1. Kinson L, Inman K. Continuous glucose monitoring in individuals with type 2 diabetes: A quality improvement program. Clin Diabetes. 2025;43(1):139-147. doi:10.2337/cd24-0006
2. Shah VN, DuBose SN, Li Z, et al. Continuous Glucose Monitoring Profiles in Healthy Nondiabetic Participants: A Multicenter Prospective Study [published correction appears in J Clin Endocrinol Metab. 2022 Mar 24;107(4):e1775-e1776. doi: 10.1210/clinem/dgab837.]. J Clin Endocrinol Metab. 2019;104(10):4356-4364. doi:10.1210/jc.2018-02763
3. Battelino T, Lalic N, Hussain S, et al. The use of continuous glucose monitoring in people living with obesity, intermediate hyperglycemia or type 2 diabetes. Diabetes Res Clin Pract. 2025;223:112111. doi:10.1016/j.diabres.2025.112111

To better understand the relationship between specific HbA1c values and cardiovascular event risk in diabetics, we studied 852,243 patients with type 2 diabetes and no prior history of stroke or MI. Time-weighted average HbA1c levels were tracked for three years starting at least one year post-diagnosis and monitored for evidence of new stroke or MI. We accounted for patient demographics, smoking status, BMI classifications, social vulnerability, and comorbidities in our analysis.

We found that the likelihood of MI rose progressively at higher average HbA1c levels, particularly above 7.0%, as seen in Figure 1. Patients with an average HbA1c between 7.0 and 7.4% were 13% more likely to experience an MI, and those with an average HbA1c of 10.0% or higher were 153% more likely, compared to patients with an average HbA1c between 6.75 and 6.99%. Notably, average HbA1c levels below 5.4% did not confer any additional benefit, with MI risk similar to or higher than for those with an average HbA1c between 6.75 and 6.99%.

The pattern for stroke was similar but less steep, with a significant rise beginning at average HbA1c levels above 7.0%. Patients are 140% more likely to have a stroke if their average HbA1c was 10.0% or higher compared to patients with an average HbA1c between 6.75 and 6.99%. The likelihood of stroke was 5% lower among patients with an average HbA1c between 6.1 and 6.4%, while the likelihood was 7% lower for those with an average HbA1c between 5.7 and 6.0%. Notably, average HbA1c levels below 5.7% did not confer any additional benefit, with stroke risk similar to or higher than for those with an average HbA1c between 6.75 and 6.99%.

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson.

1. American Diabetes Association Professional Practice Committee. 6. Glycemic goals and hypoglycemia: Standards of care in diabetes-2024. Diabetes Care. 2024;47(Suppl 1):S111-S125. doi:10.2337/dc24-S006

To understand the effects of weight cycling, we studied 156,096 adult patients with an obese BMI who had at least three BMI measurements over an 18-month period that demonstrated specific patterns in weight change. Patients were categorized into three weight patterns: sustained loss (≥10% reduction that was not regained), steady (weight remained within ±2% of baseline throughout), and regain (initial ≥10% reduction followed by return near baseline). LDL values were compared from baseline result around the time of their first qualifying BMI to an LDL result around the third BMI measurement. Patients were matched based on age, sex, BMI classification, social vulnerability, comorbidities, and use of lipid-lowering medications. 

We found that the average change in LDL differed by weight pattern. Patients in the sustained loss group had the largest average reduction, with LDL levels decreasing by 6.2 mg/dL. This contrasts with only a 1.9 mg/dL reduction in the regain group, despite similar initial weight loss, suggesting that cholesterol benefits attenuate quickly when weight is regained. Patients who maintained steady weight experienced a modest average decline of 2.7 mg/dL. The average LDL for all groups decreased, including those with stable weight, which might reflect underlying characteristics of the study population, such as increased likelihood of lipid management among those with repeat LDL testing. However, even within this potentially health-engaged subgroup, the gradient of LDL improvement across weight patterns suggests that the observed differences remain clinically meaningful. 

The distribution of LDL changes further contextualizes these average effects. Patients in the sustained loss group were more likely to experience large reductions in LDL, with a greater proportion achieving improvements of 10 mg/dL or more compared to the other groups: 41% of the loss group had LDL reductions of at least 10 mg/dL, compared to 33% in the steady group and 34% in the regain group. Patients in the regain group had a greater proportion experiencing LDL increases. Approximately 32% of regain patients had LDL increases of 10 mg/dL or more, compared to 29% of the steady group and 27% of the loss group. Previous research has shown that weight loss can lead to transient increases in LDL levels, which might be a factor contributing to the LDL increase among the sustained weight loss group.^3,4 

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson. 

1. Bays HE, Kirkpatrick CF, Maki KC, et al. Obesity, dyslipidemia, and cardiovascular disease: A joint expert review from the Obesity Medicine Association and the National Lipid Association 2024. J Clin Lipidol. 2024;18(3):e320-e350. doi:10.1016/j.jacl.2024.04.001 
2. Kakinami L, Knäuper B, Brunet J. Weight cycling is associated with adverse cardiometabolic markers in a cross-sectional representative US sample. J Epidemiol Community Health. 2020;74(8):662-667. doi:10.1136/jech-2019-213419 
3. Surampudi, V., Biggs, K., & Li, Z. (2019). Weight loss and transient LDL increase (P12-057-19). Current Developments in Nutrition, 3(Suppl 1), nzz035.P12-057-19. https://doi.org/10.1093/cdn/nzz035.P12-057-19 
4. Phinney, S. D., Tang, A. B., Waggoner, C. R., Tezanos-Pinto, R. G., & Davis, P. A. (1991). The transient hypercholesterolemia of major weight loss. The American Journal of Clinical Nutrition, 53(6), 1404–1410. https://doi.org/10.1093/ajcn/53.6.1404 

To better understand the changing landscape of endometriosis diagnoses, we studied over 25 million women 15 years of age and older who had at least one outpatient face-to-face encounter between 2017 and 2024 and determined the annual rate of new endometriosis diagnoses. Between 2017 and 2024, the annual rate of endometriosis diagnoses increased by 32%, from 24.9 to 32.8 per 10,000 patients, as seen in Figure 1. Age-specific trends show consistently higher diagnosis rates among women aged 35 to 49, who reached a rate of 85.4 per 10,000 in 2024. Diagnosis rates among women aged 25 to 34 also increased moderately, while rates in younger women (15 to 24) remained lower. 

When we evaluated symptoms in the two years leading up to the endometriosis diagnosis, we found that abdominal or pelvic pain was the most frequent, affecting 55.9% of patients who were diagnosed in 2024, as seen in Figure 2. Neuropsychiatric symptoms, including anxiety, headache, dizziness, and depression, affected nearly half of women, while menstrual problems affected 44.8%. Dyspareunia (6.3%) and infertility (3.4%) were relatively uncommon. 

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson. 

1. Zondervan, K. T., Becker, C. M., & Missmer, S. A. (2020). Endometriosis. The New England Journal of Medicine, 382(13), 1244–1256. https://doi.org/10.1056/NEJMra1810764 
2. Seo H, Burkett KM, Okocha M, et al. Social media activism and women’s health: Endometriosis awareness and support. Digit Health. 2025;11:20552076251314905. Published 2025 Jan 21. doi:10.1177/20552076251314905 

We examined data from 11 million patients with medically attended influenza infections recorded between August 1, 2013, and April 19, 2025. An influenza infection was identified by a positive lab test or a clinical diagnosis of flu. We then looked for a flu-related hospitalization or emergency department (ED) visit within the following 30 days, which were defined as those with an influenza diagnosis or a secondary condition commonly associated with an influenza infection, including COPD, heart failure, and acute necrotizing hemorrhagic encephalopathy. 

Over the past 12 flu seasons, the rate of flu-related ED visits has been the highest in the three most recent seasons. In 2022/2023, 33.1% of patients with influenza visited the ED within 30 days of the infection. This increased slightly to 35.9% in 2023/2024 and remained elevated at 35.0% in 2024/2025. For comparison, just 24.1% of patients visited the ED within 30 days of infection in the 2012/2013 season, highlighting the substantial rise in acute care use in recent years. 

Similarly, the proportion of patients hospitalized within 30 days of a flu infection has risen in recent seasons. In the 2022/2023 season, 4.9% of patients were hospitalized within 30 days of a flu infection. This increased to 5.6% in 2023/2024 and further to 7.0% in 2024/2025.  

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 299 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson. 

1. Preliminary estimated flu disease burden 2024-2025 flu season. U.S. Centers for Disease Control and Prevention. March 28, 2025. https://www.cdc.gov/flu-burden/php/data-vis/2024-2025.html. Accessed April 3, 2025. 

To further understand the relationship between childhood antibiotics and obesity, we studied 483,290 children, looking at their antibiotic courses prescribed before age 5 and their BMI at age 6. We considered patient demographics, childhood infections, diabetes, maternal obesity, and regional and socioeconomic variables in our analysis. Children who received antibiotics in the hospital were excluded. 

We found that children who received 1 to 4 antibiotic courses before age 5 were not significantly more likely to be obese at age 6 compared to those who received no antibiotics before age 5, as seen in Figure 1. Conversely, children who received 5 to 9 antibiotic courses before age 5 were 7% more likely to be obese at age 6, and those with 10 or more prescriptions were 14% more likely.  

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,700 hospitals and more than 40,000 clinics from all 50 U.S. states, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson. 

1. Trasande L, Blustein J, Liu M, Corwin E, Cox LM, Blaser MJ. Infant antibiotic exposures and early-life body mass. Int J Obes (Lond). 2013;37(1):16-23. doi:10.1038/ijo.2012.132   
2. Bailey LC, Forrest CB, Zhang P, Richards TM, Livshits A, DeRusso PA. Association of antibiotics in infancy with early childhood obesity. JAMA Pediatr. 2014;168(11):1063-1069. doi:10.1001/jamapediatrics.2014.1539