Medical Debt in the US, 2009-2020

Raymond Kluender, PhD¹; Neale Mahoney, PhD^2,3; Francis Wong, PhD³; et alWesley Yin, PhD^3,4

Author Affiliations Article Information

JAMA. 2021;326(3):250-256. doi:10.1001/jama.2021.8694

 Original Investigation

July 20, 2021

editorial comment icon

Key Points

Question  What is the total amount and distribution of medical debt in collections in the US?

Findings  In this retrospective analysis of credit reports for a nationally representative 10% panel of individuals, an estimated 17.8% of individuals in the US had medical debt in collections in June 2020 (reflecting care provided prior to the COVID-19 pandemic). Medical debt was highest among individuals who lived in the South and in zip codes in the lowest income deciles and became more concentrated in lower-income communities in states that did not expand Medicaid.

Meaning  This study provides an estimate of the amount of medical debt in collections in the US based on consumer credit reports from January 2009 to June 2020, reflecting care delivered prior to the COVID-19 pandemic, and suggests that the amount of medical debt was highest among individuals living in the South and in lower-income communities, although further study is needed regarding debt related to COVID-19.

Abstract

Importance  Medical debt is an increasing concern in the US, yet there is limited understanding of the amount and distribution of medical debt, and its association with health care policies.

Objective  To measure the amount of medical debt nationally and by geographic region and income group and its association with Medicaid expansion under the Affordable Care Act.

Design, Setting, and Participants  Data on medical debt in collections were obtained from a nationally representative 10% panel of consumer credit reports between January 2009 and June 2020 (reflecting care provided prior to the COVID-19 pandemic). Income data were obtained from the 2014-2018 American Community Survey. The sample consisted of 4.1 billion person-month observations (nearly 40 million unique individuals). These data were used to estimate the amount of medical debt (nationally and by geographic region and zip code income decile) and to examine the association between Medicaid expansion and medical debt (overall and by income group).

Exposures  Geographic region (US Census region), income group (zip code income decile), and state Medicaid expansion status.

Main Outcomes and Measures  The stock (all unpaid debt listed on credit reports) and flow (new debt listed on credit reports during the preceding 12 months) of medical debt in collections that can be collected on by debt collectors.

Results  In June 2020, an estimated 17.8% of individuals had medical debt (13.0% accrued debt during the prior year), and the mean amount was $429 ($311 accrued during the prior year). The mean stock of medical debt was highest in the South and lowest in the Northeast ($616 vs $167; difference, $448 [95% CI, $435-$462]) and higher in poor than in rich zip code income deciles ($677 vs $126; difference, $551 [95% CI, $520-$581]). Between 2013 and 2020, the states that expanded Medicaid in 2014 experienced a decline in the mean flow of medical debt that was 34.0 percentage points (95% CI, 18.5-49.4 percentage points) greater (from $330 to $175) than the states that did not expand Medicaid (from $613 to $550). In the expansion states, the gap in the mean flow of medical debt between the lowest and highest zip code income deciles decreased by $145 (95% CI, $95-$194) while the gap increased by $218 (95% CI, $163-$273) in the nonexpansion states.

Conclusions and Relevance  This study provides an estimate of the amount of medical debt in collections in the US based on consumer credit reports from January 2009 to June 2020, reflecting care delivered prior to the COVID-19 pandemic, and suggests that the amount of medical debt was highest among individuals living in the South and in lower-income communities. However, further study is needed regarding debt related to COVID-19.

Introduction

Due to rising health care prices,^1,2 increased cost sharing,³ and 26.1 million individuals without insurance in 2019,⁴ the US health care system leaves patients with high out-of-pocket costs.^1,5 If these medical bills are unpaid, the outstanding amount can be classified as medical debt and sent to debt collectors.

Medical debt is associated with reduced health care use.⁶ Personal debt, broadly defined, is associated with worse mental health^7,8 and a deterioration of personal finances.⁹ Despite widespread concern, there is only limited evidence on recent trends in medical debt, its distribution across individuals, and how health policy has affected the distribution of medical debt. To our knowledge, no studies have estimated the total amount of medical debt. Moreover, even though recent studies have estimated the effect of the Affordable Care Act (ACA) coverage expansion on individual consumer debt,^9,10 no study has investigated its association with differences across geographic areas and socioeconomic groups.

To study these issues, a nationally representative person-level 10% sample of all consumer credit reports observed monthly from 2009 through 2020 was used to (1) document the scale and prevalence of medical debt nationally, (2) characterize differences in medical debt across geographic regions and income groups, and (3) examine the association between ACA Medicaid expansion and the distribution of medical debt across individuals.

Methods

Data Sources and Study Population

We measured unpaid medical debt in collections using a nationally representative, randomly selected 10% panel of all individuals with credit reports maintained by TransUnion, which is 1 of the 3 nationwide credit reporting agencies. The data were deidentified and included birth dates, zip codes, loan repayment history, public records, and medical and nonmedical accounts in collections. Because the data were not specifically collected for this study and were deidentified, the study was not considered human subjects research.¹¹

Persons in the credit panel were monitored each month from January 2009 through June 2020. We excluded persons with a missing age or zip code, those residing outside the 50 states or the District of Columbia, and those with empty reports (defined as reports with no credit records of any kind). We also replicated the analyses using a sample that retains these empty accounts (eMethods, eFigures 1-4, and eTable 1 in the Supplement). Additional details on the sample construction appear in the eMethods in the Supplement.

Medical debt is reported to TransUnion by third-party debt collectors. It is typically reported at least 180 days after the bill was incurred and must be removed from the credit report after 7 years.¹² For each medical debt item, we observed the date when the bill became delinquent, its current balance, and whether the debt was in dispute or had been closed. We excluded medical debts that had been paid, were in dispute, or that had been closed. Dollar amounts were adjusted for inflation to June 2020 using the consumer price index for all urban consumers and censored at the 99.99th percentile to reduce the influence of extreme outliers.

Exposures

We documented regional patterns by reporting measures of medical debt by US Census region (a list of states by US Census region appears in eTable 2 in the Supplement) and by constructing county-level maps of mean medical debt.

To analyze differences in medical debt by income, we computed the mean stock (all unpaid debt listed on credit reports) of medical debt by zip code income decile. We assigned each zip code to a decile using per-capita income estimates from the 5-year American Community Survey (2014-2018), weighting each zip code by its population in the survey. We then calculated the measures of medical debt separately for each decile group.

To analyze the association between medical debt and Medicaid expansion, we assigned states to 1 of 3 groups: (1) states that expanded Medicaid in 2014 (28 states), (2) states that expanded Medicaid after 2014 (11 states); and (3) states that did not expand Medicaid (12 states). Except for a small number of states with limited early expansion, Medicaid expansion was first implemented in 2014.¹³ A list of states by Medicaid expansion status appears in eTable 3 in the Supplement.

Outcomes

The main outcomes were the stock (defined above) and flow (new debt listed on credit reports during the preceding 12 months) of medical debt. The stock is the preferred measure of total medical debt.

Because debt in collections can remain on credit reports for 7 years, it can take up to 7 years for changes in policy to be fully reflected in the stock of medical debt. To analyze the association between Medicaid expansion under the ACA and medical debt, we constructed a measure of the flow of medical debt. The flow measure is unaffected by outflows (arising from repayment or because a collector stops reporting on the debt) and thus can be used to assess the sensitivity of the estimates to the potential differences in outflows across industries, regions, or income groups.

To maintain a consistent reference point, we computed statistics as of June for each year. Because of the 180-day delay in credit reporting, statistics from June 2020 (the most recent month in the study sample) did not reflect medical care received by patients in 2020 and were unaffected by the COVID-19 pandemic.

For comparison, we constructed analogous measures of nonmedical debt in collections. Nonmedical debt combines all other sources of debt in collections, including credit cards, personal loans, utilities, and phone bills. As with medical debt, these amounts represent bills that are unpaid and can be collected on by debt collectors.

Statistical Analysis

To analyze the association between medical debt and Medicaid expansion, we conducted a difference-in-differences analysis that compared percentage changes over time in the flow of medical debt in states that expanded Medicaid vs the percentage changes over time in states that did not expand Medicaid. Specifically, we normalized the level of medical debt for each group of states to 1 in 2013 (the year before Medicaid expansion in the intervention group) and calculated the percentage change relative to this year. This normalization adjusts for any cross-sectional differences among groups of states that would otherwise confound the estimates. The trends without this normalization appear in eFigure 5 in the Supplement.

In the sensitivity analysis, we estimated the association between medical debt and Medicaid expansion with linear regressions of the percentage change in the mean flow of medical debt between 2013 and 2020 on indicators for the Medicaid expansion groups and a constant, controlling for state-level changes in economic factors (unemployment rate, the percentage of the population aged ≥65 years, the percentage of individuals aged ≥25 years and with a bachelor’s degree or higher, and median income), the state-level share of beds at for-profit hospitals, and state-level policies (debt collection laws and surprise out-of-network billing laws). For some of the control variables, data for 2020 were not available, so we used data from 2013 and 2019. Additional details appear in eTables 4-5 in the Supplement.

To assess whether the association between Medicaid expansion and medical debt reflected confounding factors (such as differential economic trends), we conducted the analyses separately using nonmedical debt as the outcome. To analyze the relationship between Medicaid expansion and income-based differences in medical debt, we estimated the mean flow of medical debt between 2009 and 2020 by zip code income decile separately for Medicaid expansion and nonexpansion states. To maintain consistent zip code income deciles across Medicaid expansion and nonexpansion states, we continued to assign zip codes to income deciles based on their population-weighted rank in the national distribution.

The raw data were extracted and collapsed to the zip code year using Spark SQL version 2.2.1 (Apache Software Foundation) via Python version 3.6.3 (Python Software Foundation). The processed data were analyzed using Stata/MP version 16.0 (StataCorp). The 95% CIs for the national, regional, and zip code income decile estimates were based on standard errors constructed using the zip code year data. Tests of statistical significance were based on 2-sided tests with a significance threshold of .05. Because of the potential for type I error due to multiple comparisons, the findings for the secondary analyses should be interpreted as exploratory.

Results

National Trends in the Mean Stock and Flow of Medical Debt

The 10% panel of consumer credit reports held 4.1 billion person-month observations and nearly 40 million (n = 39 788 671) unique individuals. The mean stock of medical debt increased from $750 in 2009 to a peak of $827 in 2010 (difference, $76 [95% CI, $61-$92]) before decreasing to $429 in 2020 (difference, $397 [95% CI, $384-$411]) (Figure 1A). During this period, medical debt overtook nonmedical debt as the largest source of debt in collections. In 2009, mean medical debt was $119 (95% CI, $112-$127) less than nonmedical debt, whereas in 2020 medical debt exceeded nonmedical debt by $39 (95% CI, $34-$43). Numerical values for each year appear in eTable 6 in the Supplement.

The mean flow of medical debt also increased from $332 in 2009 to a peak of $427 in 2013 (difference, $95 [95% CI, $87-$103]) before decreasing to $311 in 2020 (difference, $116 [95% CI, $108-$124]) (Figure 1B). Similar to the stock, the flow of medical debt overtook nonmedical debt as the largest source of new debt in collections during this period. In 2009, medical debt was $120 (95% CI, $116-$124) less than nonmedical debt, whereas in 2020 medical debt exceeded nonmedical debt by $78 (95% CI, $75-$82).

Summary statistics on the amount of US debt in 2020 and by US Census region appear in the Table. At the national level, 17.8% of persons with a credit report had medical debt in collections and 13.0% accrued medical debt during the prior year. Conditional on having medical debt, the mean stock was $2424 and the mean flow was $2396.

Regional and Income-Based Differences in Medical Debt

The mean stock of medical debt by county in 2020 appears in Figure 2A (the analogous map for the mean flow of medical debt by county appears in eFigure 6A in the Supplement). Medical debt was concentrated in the South as well as in some counties in the West. Of the 4 US Census regions, the South had the highest amount of medical debt, with medical debt held by 23.8% of persons and a mean stock of $616 (95% CI, $608-$623). The Northeast had the lowest amount of medical debt, with medical debt held by 10.8% of persons and a mean stock of $167 (95% CI, $163-$172). The difference between the South and Northeast was $448 (95% CI, $435-$462).

The mean stock of medical debt by zip code income decile appears in Figure 2B (the numerical values appear in eTable 7 and the statistics for the mean flow of medical debt by zip code income decile appear in eFigure 6B in the Supplement). The mean stock of medical debt was $677 (95% CI, $662-$692) in the 1st (lowest) zip code income decile, $473 (95% CI, $462-$484) in the 5th zip code income decile, and $126 (95% CI, $121-$131) in the 10th (highest) zip code income decile. The difference between the lowest and highest zip code deciles was $551 (95% CI, $520-$581).

Medicaid Expansion Under the ACA and Accrual of Medical Debt

The flow of medical debt between 2009 and 2020 by Medicaid expansion status appears in Figure 3A, with the levels for each group of states normalized to 1 in 2013. Between 2013 and 2020, the states that expanded Medicaid in 2014 experienced a decline in the mean flow of medical debt that was 34.0 percentage points (95% CI, 18.5-49.4 percentage points) greater (from $330 to $175) than the states that did not expand Medicaid (from $613 to $550). States that expanded Medicaid after 2014 experienced a decline in the mean flow of medical debt that was 20.4 percentage points (95% CI, 1.2-39.6 percentage points) greater (from $401 to $288) than the states that did not expand Medicaid (from $613 to $550). The numerical values appear in eTable 8 in the Supplement. The regression tables that underlie these estimates show consistent results from the specifications that control for economic and policy factors (eTables 4-5 in the Supplement).

The analogous plots for nonmedical debt appear in Figure 3B and the estimates from the regression specifications appear in eTable 9 in the Supplement. For nonmedical debt, there were no significant differences across the groups. Expansion states experienced a decline in the mean flow of nonmedical debt of 45.4% (95% CI, 38.8%-52.1%) from $388 to $206; late expansion states experienced a decline of 37.7% (95% CI, 33.0%-42.4%) from $337 to $212; and nonexpansion states experienced a decline of 40.9% (95% CI, 35.3%-46.6%) from $488 to $287.

Medicaid Expansion Under the ACA and Income-Based Differences in the Accrual of Medical Debt

The flow of medical debt in 2009 and 2020 by zip code income decile appears separately for states that expanded Medicaid (Figure 4A) and for states that did not expand Medicaid (Figure 4B). The states that expanded Medicaid after 2014 are not shown.

Within states that expanded Medicaid (Figure 4A), all zip code income deciles experienced reductions in medical debt from 2009 to 2020, with larger reductions in the lower zip code income deciles. In the lowest zip code income decile, the mean flow of medical debt decreased by $180 (95% CI, $131-$229) from $458 to $278. In the highest zip code income decile, the mean flow of medical debt decreased by $35 (95% CI, $27-$44) from $95 to $60. The gap in the mean flow of medical debt between the lowest and highest zip code income deciles decreased by $145 (95% CI, $95-$194) from $363 to $218.

Within states that did not expand Medicaid (Figure 4B), most zip code income deciles experienced an increase in medical debt from 2009 to 2020; there were greater increases in the lowest zip code income deciles. In the lowest zip code income decile, the mean flow of medical debt increased by $206 (95% CI, $156-$256) from $630 to $836. In the highest zip code income decile, the mean flow of medical debt decreased by $12 (95% CI, $10-$35) from $196 to $184. The gap in the mean flow of medical debt between the lowest and highest zip code income deciles increased by $218 (95% CI, $163-$273) from $434 to $652.

Discussion

In a retrospective analysis of consumer credit reports, the mean amount of medical debt was high, and it was greater among individuals who lived in the South and in zip codes in the lowest income deciles. Medicaid expansion under the ACA was associated with reduced medical debt overall, and with reduced gaps in the amount of medical debt between low-income and high-income communities.

During the last decade, medical debt has become the largest source of debt in collections. The reductions in nonmedical debt in collections between 2009 and 2020 occurred simultaneously with the economic recovery from the Great Recession, consistent with the well-documented association between unemployment and loan delinquency.¹⁴ In contrast, total medical debt in collections decreased by a more modest amount. As a result, as of June 2020 individuals had $39 more in mean medical debt in collections than they had in mean debt in collections from all other sources combined ($429 vs $390), including credit cards, utilities, and phone bills.

The randomized sampling design can be used to extrapolate to the total amount of medical debt in collections reported to TransUnion. Multiplying mean medical debt of $429 in June 2020 by the sample size and a factor of 10 (because the sample is a 10% random sample of TransUnion credit reports) implies the existence of $140 billion in total medical debt. Even though this extrapolation can estimate total medical debt in collections listed on TransUnion credit reports, the limitations of the data preclude extrapolation to total medical debt in the US.

The analysis shows that Medicaid expansion was associated with reductions in medical debt in collections. Although the study design does not allow for causal interpretation, the absence of a meaningful association between Medicaid expansion and changes in nonmedical debt, and the stability of the estimates, controlling for economic and policy factors, reduce concerns about possible confounders.¹⁵ These estimates are consistent with studies that have used experimental methods to establish a causal link between Medicaid coverage and reductions in medical debt.¹⁶

Many of the states with the highest pre-ACA levels of medical debt did not expand Medicaid and subsequently did not experience substantial reductions in medical debt. Specifically, 8 of the 12 states that did not expand Medicaid are in the South, the region with the highest pre-ACA levels of medical debt (eTable 3 in the Supplement). The mean medical debt decreased by 44.0% between 2013 and 2020 in states that immediately expanded Medicaid, but it only decreased by 10.0% in the nonexpansion states, exacerbating preexisting regional differences.

Within the states that expanded Medicaid, there were reductions in income-based differences. In states that expanded Medicaid, the gap in the flow of medical debt between those living in in the lowest and highest zip code income deciles decreased over the 2009 to 2020 period, whereas the gap in the flow of medical debt increased in states that did not expand Medicaid.

Taken together, the results on income and regional differences indicate that individuals in the lowest zip code income deciles in states that did not expand Medicaid had the highest levels of medical debt in the country at the start of the study period and also experienced the largest subsequent increases in mean medical debt. For instance, the lowest zip code income decile in states that did not expand Medicaid had both the highest level of medical debt in 2009 and the largest increase over the 2009 to 2020 period.

Limitations

This study has several limitations. First, the measures of medical debt were derived from debts that were reported to TransUnion, which may not be identical to the debts reported to other credit bureaus.¹⁷

Second, the data did not allow for the measurement of medical debts that are not reported to credit bureaus, which may differ by industry, region, or income group. Third, the data may have included reports for persons who had emigrated or multiple reports for a person that were not linked.

Fourth, the stocks of debt in collection (but not the flows) were influenced by outflows from credit reports; outflows may differ by industry, region, and income group and are difficult to interpret. Fifth, the measures of medical debt did not capture medical expenses that were paid with a credit card or other financial products.

Sixth, even though the study examined the association between medical debt and income at the zip code level, it was not possible to study this (or other associations) at the individual level. Seventh, the regression specifications that examined the association between Medicaid expansion and medical debt were unable to control for individual-level factors or all potentially relevant time-varying state-level confounders.

Eighth, because medical debt is reported to credit bureaus after a 180-day delay, the measures of medical debt in 2020 did not capture debt incurred during the COVID-19 pandemic (or any debt from care provided in 2020).

Conclusions

This study provides an estimate of the amount of medical debt in collections in the US based on consumer credit reports from January 2009 to June 2020, reflecting care delivered prior to the COVID-19 pandemic, and suggests that the amount of medical debt was highest among individuals living in the South and in lower-income communities. However, further study is needed regarding debt related to COVID-19.

Back to top

Article Information

Corresponding Author: Neale Mahoney, PhD, Stanford University, 579 Jane Stanford Way, Stanford, CA 94305 (nmahoney@stanford.edu).

Accepted for Publication: May 13, 2021.

Author Contributions: Dr Mahoney had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: All authors.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: All authors.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: All authors.

Administrative, technical, or material support: Yin.

Supervision: Mahoney, Yin.

Conflict of Interest Disclosures: None reported.

Funder/Support: Dr Mahoney was supported by internal research funds from the University of Chicago (where he was a faculty member through June 2020) and Stanford University (where he has been a faculty member since July 2020). Dr Wong was supported by grant T32-AG000186 from the National Institute on Aging. The credit data used in this study were provided by TransUnion, a global information solutions company, through a relationship with the Kilts Center for Marketing at the University of Chicago Booth School of Business.

Role of the Funder/Sponsor: No funder/sponsor had any role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation of the manuscript; and decision to submit the manuscript for publication. TransUnion had the right to review the research before dissemination to ensure it accurately describes TransUnion data, does not disclose confidential information, and does not contain material it deems to be misleading or false regarding TransUnion, TransUnion’s partners, affiliates or customer base, or the consumer lending industry.

Additional Contributions: We thank Xuyang Xia, BA (research assistant at Stanford University), for making substantial contributions to the data analysis and who was compensated.

References

1.

Health Care Cost Institute. 2017 health care cost and utilization report. Accessed March 14, 2020. https://www.healthcostinstitute.org/images/pdfs/HCCI_2017_%20Health_%20Care_Cost_and_Utilization_Report_02.12.19.pdf

2.

Dieleman  JL, Squires  E, Bui  AL,  et al.  Factors associated with increases in US health care spending, 1996-2013.   JAMA. 2017;318(17):1668-1678.
ArticlePubMedGoogle ScholarCrossref

3.

Kaiser Family Foundation. 2018 employer health benefits survey. Accessed March 14, 2020. https://www.kff.org/report-section/2018-employer-health-benefits-survey-section-7-employee-cost-sharing/attachment/figure-7-14/

4.

Keisler-Starkey  K, Bunch  LN.  Health insurance coverage in the United States: 2019.  Accessed March 14, 2020. https://www.census.gov/content/dam/Census/library/publications/2020/demo/p60-271.pdf

5.

Centers for Medicare & Medicaid Services. National health expenditure fact sheet. Accessed March 14, 2020. https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/NationalHealthExpendData/NHE-Fact-Sheet

6.

Kalousova  L, Burgard  SA.  Debt and foregone medical care.   J Health Soc Behav. 2013;54(2):204-220.PubMedGoogle ScholarCrossref

7.

Jenkins  R, Bhugra  D, Bebbington  P,  et al.  Debt, income and mental disorder in the general population.   Psychol Med. 2008;38(10):1485-1493.PubMedGoogle ScholarCrossref

8.

Meltzer  H, Bebbington  P, Brugha  T,  et al.  The relationship between personal debt and specific common mental disorders.   Eur J Public Health. 2013;23(1):108-113.PubMedGoogle ScholarCrossref

9.

Brevoort  K, Grodzicki  D, Hackmann  MB.  The credit consequences of unpaid medical bills.   J Public Econ. 2020;187:104203. doi:10.1016/j.jpubeco.2020.104203Google Scholar

10.

Hu  L, Kaestner  R, Mazumder  B, Miller  S, Wong  A.  The effect of the Affordable Care Act Medicaid expansions on financial wellbeing.   J Public Econ. 2018;163:99-112.PubMedGoogle ScholarCrossref

11.

Protection of human subjects, 45 CFR Part 690. Accessed June 15, 2020. https://www.govinfo.gov/app/details/CFR-2012-title45-vol3/CFR-2012-title45-vol3-part690

12.

Requirements relating to information contained in consumer reports, 15 US Code §1681c. Accessed June 15, 2020. https://www.govinfo.gov/app/details/USCODE-2010-title15/USCODE-2010-title15-chap41-subchapIII-sec1681c

13.

Sommers  BD, Kenney  GM, Epstein  AM.  New evidence on the Affordable Care Act.   Health Aff (Millwood). 2014;33(1):78-87.PubMedGoogle ScholarCrossref

14.

Gerardi  K, Herkenhoff  KF, Ohanian  LE, Willen  PS.  Can’t pay or won’t pay?   Rev Financ Stud. 2018;31(3):1098-1131. doi:10.1093/rfs/hhx115Google ScholarCrossref

15.

Oster  E.  Unobservable selection and coefficient stability.   JBES. 2019;37(2):187-204. doi:10.1080/07350015.2016.1227711Google Scholar

16.

Finkelstein  A, Taubman  S, Wright  B,  et al.  The Oregon health insurance experiment.   Q J Econ. 2012;127(3):1057-1106.PubMedGoogle ScholarCrossref

17.

Federal Trade Commission. Report to Congress under section 319 of the Fair and Accurate Credit Transactions Act of 2003. Accessed March 14, 2020. https://www.ftc.gov/sites/default/files/documents/reports/section-319-fair-and-accurate-credit-transactions-act-2003-fifth-interim-federal-trade-commission/130211factareport.pdf

PRIVATE MEDICINE U$A

Antibiotic prescriptions for kids plummet during pandemic

Study finds declines in prescription drugs dispensed to children during COVID-19, including infection-related medicines and some used for chronic diseases

MICHIGAN MEDICINE - UNIVERSITY OF MICHIGAN

Research News

ANN ARBOR, Mich. - As children made fewer visits to health facilities and engaged in social distancing and other COVID-19 mitigation measures, a smaller number of them also received prescription drugs, a new study suggests.

Overall, medications prescribed for children dropped by more than a quarter during the first eight months of the pandemic compared to the previous year, with the steepest declines in infection-related medicines like antibiotics and cough-and-cold drugs.

Antibiotic dispensing to children and teens plunged by nearly 56 % between April and December 2020 compared with the same period in 2019. Researchers also found declines in prescriptions for chronic diseases, such as attention deficit hyperactivity disorder (ADHD) and asthma, but no change in prescriptions for antidepressants, according to the findings in Pediatrics.

"The decline in the number of children receiving antibiotics is consistent with the large decreases in infection-related pediatric visits during 2020," said lead author Kao-Ping Chua, M.D., Ph.D., a pediatrician and researcher at University of Michigan Health C.S. Mott Children's Hospital and the Susan B. Meister Child Health Evaluation and Research Center.

"Because antibiotics have important side effects, the dramatic decreases in antibiotic dispensing may be a welcome development," he added. "However, declines in dispensing of chronic disease drugs could be concerning."

Dispensing of infection-related drugs declined sharply

Researchers analyzed national prescription drug dispensing data from 92% of U.S pharmacies to assess changes in dispensing to children ages 0-19 during COVID-19.

Between January 2018 and February 2020, nearly 25.8 million prescriptions were dispensed to children a month. Dispensing totals during the first 8 months of the pandemic dropped by about 27% compared to the same period in 2019.

Overall, drugs typically prescribed for acute infections, including antibiotics, fell by nearly 51 % while those for chronic diseases fell by 17 %.

"The decrease in antibiotic dispensing most likely reflects reductions in infections, such as colds and strep throat, due to COVID-19 risk-mitigation measures like social distancing and face masks," Chua said.

"As a result, children had fewer infection-related visits and had fewer opportunities to receive antibiotic prescriptions, whether for antibiotic-appropriate conditions or antibiotic-inappropriate conditions."

Chua's previous research has suggested that nearly a quarter of antibiotic prescriptions among children and adults may be unnecessary. In children, antibiotics are the leading cause of emergency room visits for adverse drug events, with potential side effects including allergic reactions, fungal infections and diarrhea.

Long term, antibiotic overuse may also contribute to antibiotic-resistant bacteria development, causing illnesses that were once easily treatable with antibiotics to become untreatable and dangerous, Chua said.

Another welcomed development in drug dispensing trends, researchers found, was a decline in prescription medicines to treat symptoms of the common cold, particularly to suppress coughs. Findings suggest a nearly 80 % drop in these medications (known as antitussive drugs) during the 2020 study period.

"These drugs have little benefit but are associated with potentially harmful side effects, particularly in young children," Chua said.

"From the perspective of health care quality, the sharp decline in dispensing of cough-and-cold medications may represent a silver lining of the COVID-19 pandemic."

While dispensing of infection-related drugs to children could rebound as social distancing measures are lifted and infections increase, it may still not necessarily return to pre-pandemic levels soon, Chua said. If COVID-19 risk-mitigation measures continue in schools and day cares, for example, this may lower the incidence of conditions for which antibiotics are frequently prescribed, such as ear infections, sinusitis, and upper respiratory infections.

Dispensing of chronic disease drugs

The study found a modest 11% decline in dispensing of prescriptions for ADHD.

"Whether this decline is concerning needs to be studied further," Chua said. "For example, it is unclear whether the decline in ADHD prescriptions reflect a reduced need for medications at school due to the transition to remote learning, disruptions in medication access, or delays in diagnosis."

There were also large declines in dispensing of asthma medications, such as albuterol and inhaled steroids, according to the research

National data suggest that the number of asthma attacks in children has dropped sharply during the pandemic, Chua said. Given this, the decline in medication dispensing likely reflects better control of asthma.

Researchers need more data to better understand the lack of change in antidepressant dispensing to children during the pandemic.

"An optimistic view is that few children on established antidepressant regimens discontinued use," Chua said.

"Studies, however, suggest that the mental health of children has worsened during the pandemic, particularly among adolescents. Given this, our findings might suggest that antidepressant dispensing has not risen to meet this increased need."

Clinicians may be able to use electronic health records to identify decreases in the frequency of refill requests among children on established drug regimens for chronic disease, Chua said. Clinicians could then call families to determine if there is a reason for concern - such as medications not being affordable for them - or if the changes reflect improved disease control.

Dispensing totals declined more sharply for prescriptions paid with cash than for other payer types. Chua believes that this finding suggests that uninsured children faced greater financial-related barriers to accessing medical care and prescription drugs during the pandemic.

The decreased dispensing in kids is consistent with the drop in total number of prescriptions dispensed to adult Americans, which declined sharply during the pandemic but subsequently rebounded. However, the study indicates dispensing to children has not rebounded to the same degree, Chua said.

"This study provides a national picture of prescription drug dispensing to children before and during the pandemic," he said. "It will be important to monitor whether the reductions we demonstrate are temporary or sustained."

###

 

Study finds surprising source of social influence

Want to promote your new product or trigger a shift in thinking? Steer clear of the influencers.

VIDEO: AS PROMINENT AND REVERED AS SOCIAL INFLUENCERS SEEM TO BE, IN FACT, THEY ARE UNLIKELY TO CHANGE A PERSON'S BEHAVIOR BY EXAMPLE -- AND MIGHT ACTUALLY BE DETRIMENTAL TO THE... view more 

CREDIT: ANNENBERG SCHOOL FOR COMMUNICATION, UNIVERSITY OF PENNSYLVANIA

Imagine you're a CEO who wants to promote an innovative new product -- a time management app or a fitness program. Should you send the product to Kim Kardashian in the hope that she'll love it and spread the word to her legions of Instagram followers? The answer would be 'yes' if successfully transmitting new ideas or behavior patterns was as simple as showing them to as many people as possible.

However, a forthcoming study in the journal Nature Communications finds that as prominent and revered as social influencers seem to be, in fact, they are unlikely to change a person's behavior by example -- and might actually be detrimental to the cause.

Why?

"When social influencers present ideas that are dissonant with their followers' worldviews -- say, for example, that vaccination is safe and effective -- they can unintentionally antagonize the people they are seeking to persuade because people typically only follow influencers whose ideas confirm their beliefs about the world," says Damon Centola, Elihu Katz Professor of Communication, Sociology, and Engineering at Penn, and senior author on the paper.

So what strategy do we take if we want to use an online or real world neighborhood network to 'plant' a new idea? Is there anyone in a social network who is effective at transmitting new beliefs? The new study delivers a surprising answer: yes, and it's the people you'd least expect to have any pull. To stimulate a shift in thinking, target small groups of people in the "outer edge" or fringe of a network.

Centola and Douglas Guilbeault, Ph.D., a recent Annenberg graduate, studied over 400 public health networks to discover which people could spread new ideas and behaviors most effectively. They tested every possible person in every network to determine who would be most effective for spreading everything from celebrity gossip to vaccine acceptance.

"Dozens of algorithms that are currently used by enterprises seeking to spread new ideas are based on the fallacy that everything spreads virally," says Centola. "But this study shows that the ability for information to pass through a social network depends on what type of information it is."

So, if you want to spread gossip -- easily digestible, uncontroversial bits of information -- go ahead and tap an influencer. But if you want to transmit new ways of thinking that challenge an existing set of beliefs, seek out hidden locations in the periphery and plant the seed there.

"Our big discovery," Centola added, "is that every network has a hidden social cluster in the outer edges that is perfectly poised to increase the spread of a new idea by several hundred percent. These social clusters are ground zero for triggering tipping points in society."

Centola and Guilbeault applied their findings to predicting the spread of a new microfinance program across dozens of communities in India. By considering what was being spread through the networks, they were able to predict where it should originate from, and whether it would spread to the rest of the population. Their predictions identified the exact people who were most influential for increasing the adoption of the new program.

Guilbeault, now an assistant professor at the University of California, Berkeley, noted, "in a sense, we found that the center of the network changed depending on what was spreading. The more uncertain people were about a new idea, the more that social influence moved to the people who only had parochial connections, rather than people with many far-reaching social connections." Guilbeault added, "the people in the edges of the network suddenly had the greatest influence across the entire community."

The findings "turn our notions about social influence for marketing, sales, and social movements upside down," says Centola. "Not everything spreads through a network in the same way," he adds, "and we can use this knowledge to pinpoint hotspots in the social graph. This can allow us to accurately tailor our network strategies for effecting positive social change."

Centola is the author of the new book, Change: How to Make Big Things Happen (Little Brown, 2021).

###

Dearth of mental health support during pandemic for those with chronic health problems

A scoping review of English and Chinese-language studies found individuals with chronic physical health conditions had higher levels of mental health problems during the COVID-19 pandemic; targeted interventions were lacking

UNIVERSITY OF TORONTO

Research News

A new scoping review found that those with chronic health concerns, such as diabetes, heart disease, cancer, and autoimmune conditions, are not only at a higher risk of severe COVID-19 infection, they are also more likely to experience anxiety, depression or substance use during the COVID-19 pandemic.

The aim of the review was to address knowledge gaps related to the prevention and management of mental health responses among those with chronic conditions. The findings, recently published in the International Journal of Environmental Research and Public Health, were based on a comprehensive review of 67 Chinese and English-language studies.

"Levels of anxiety, depression, and substance use tended to be more prevalent among those with physical health concerns, and these mental health impacts also interfered with their treatment plans," says first author Karen Davison, Canada Research Chair at Kwantlen Polytechnic University.

Physical and mental health problems often occur together, possibly due to factors such as shared underlying inflammatory responses and the psychosocial effects of living with a health condition, say the study's authors. Economic instability, social isolation, and reduced access to health and social care services also increased the likelihood of mental health concerns among those with a chronic physical health condition.

"These circumstances, which became more prevalent during the pandemic, likely impact an individual's ability to cope," says co-author Professor Simon Carroll from the University of Victoria's Sociology department.

Rapidly spreading misinformation during the pandemic may have also influenced reactions that can worsen mental health.

"Lower levels of health literacy have been associated with poorer physical and mental health," says Brandon Hey, Policy and Research Analyst, COVID 19 Policy, Programs and Priorities at the Mental Health Commission of Canada. "This needs to be addressed by the public health community who can educate and support social and conventional media to accurately deliver information."

The findings and practice recommendations from this review have the potential to inform the work of policy-makers, practitioners, and researchers looking to provide better mental health supports for those with chronic illness.

"Several promising practices include screening for mental health issues, addressing factors such as income support, using digital resources to provide care, and providing services such as patient navigation, group online visits, peer support, and social prescribing," says co-author University of British Columbia Nursing Professor Maura MacPhee.

University of Toronto Social Work Professor, Esme Fuller-Thomson, who is also Director of the Institute for Life Course and Aging, says we now have the opportunity to shape policies, programs, and other efforts to strengthen people's mental health. "Multi-integrated interventions can help provide the supports that are needed to address the complex needs of different populations and foster resilience in times of public health crises," she says.

###

The review, funded by the Canadian Institutes of Health Research, was produced by a team of researchers from universities in Canada and the UK, the Mental Health Commission of Canada and other health organizations, and patient advisors.

 

JAVACRUCIAN NEWS WORTH REPEATING  

Coffee doesn't raise your risk for heart rhythm problems

UCSF Cardiology researchers report no link between coffee consumption and arrhythmia

UNIVERSITY OF CALIFORNIA - SAN FRANCISCO

Research News

In the largest study of its kind, an investigation by UC San Francisco has found no evidence that moderate coffee consumption can cause cardiac arrhythmia.

In fact, each additional daily cup of coffee consumed among several hundred thousand individuals was associated with a 3 percent lower risk of any arrhythmia occurring, including atrial fibrillation, premature ventricular contractions, or other common heart conditions, the researchers report. The study included a four-year follow up.

The paper is published July 19, 2021, in JAMA Internal Medicine.

"Coffee is the primary source of caffeine for most people, and it has a reputation for causing or exacerbating arrhythmias," said senior and corresponding author Gregory Marcus, MD, professor of medicine in the Division of Cardiology at UCSF.

"But we found no evidence that caffeine consumption leads to a greater risk of arrhythmias," said Marcus, who specializes in the treatment of arrhythmias. "Our population-based study provides reassurance that common prohibitions against caffeine to reduce arrhythmia risk are likely unwarranted."

 

While some professional societies suggest avoiding caffeinated products to lower the risk for arrhythmia, this connection has not been consistently demonstrated - indeed, coffee consumption may have anti-inflammatory benefits and is associated with reduced risks of some illnesses including cancer, diabetes, and Parkinson disease.

In the new study, UCSF scientists explored whether habitual coffee intake was associated with a risk of arrhythmia, and whether genetic variants that affect caffeine metabolism could modify that association. Their investigation was conducted via the community-based UK Biobank, a prospective study of participants in England's National Health Services.

Some 386,258 coffee drinkers took part in the coffee research, with an average mean age of 56 years; slightly more than half were female. It was an unprecedented sample size for this type of inquiry.

In addition to a conventional analysis examining self-reported coffee consumption as a predictor of future arrhythmias, the investigators employed a technique called "Mendelian Randomization," leveraging genetic data to infer causal relationships. As those with the genetic variants associated with faster caffeine metabolism drank more coffee, this analysis provided a method to test the caffeine-arrhythmia relationship in a way that did not rely on participant self-report and should have been immune to much of the confounding inherent to most observational studies.

With a mean four-year follow up, data were adjusted for demographic characteristics, health and lifestyle habits.

Ultimately, approximately 4 percent of the sample developed an arrhythmia. No evidence of a heightened risk of arrhythmias was observed among those genetically predisposed to metabolize caffeine differently. The researchers said that higher amounts of coffee were actually associated with a 3 percent reduced risk of developing an arrhythmia.

The authors noted limitations including the self-reporting nature of the study, and that detailed information on the type of coffee - such as espresso or not - was unavailable.

"Only a randomized clinical trial can definitively demonstrate clear effects of coffee or caffeine consumption," said Marcus. "But our study found no evidence that consuming caffeinated beverages increased the risk of arrhythmia. Coffee's antioxidant and anti-inflammatory properties may play a role, and some properties of caffeine could be protective against some arrhythmias."

###

Co-authors are Eun-jeong Kim, MD; Thomas J. Hoffmann, PhD; Gregory Nah, MA; Eric Vittinghoff, PhD; and Francesca Delling, MD, all of UCSF.

Disclosures can be found in the paper.

About UCSF Health: UCSF Health is recognized worldwide for its innovative patient care, reflecting the latest medical knowledge, advanced technologies and pioneering research. It includes the flagship UCSF Medical Center, which is ranked among the top 10 hospitals nationwide, as well as UCSF Benioff Children's Hospitals, with campuses in San Francisco and Oakland, Langley Porter Psychiatric Hospital and Clinics, UCSF Benioff Children's Physicians and the UCSF Faculty Practice. These hospitals serve as the academic medical center of the University of California, San Francisco, which is world-renowned for its graduate-level health sciences education and biomedical research. UCSF Health has affiliations with hospitals and health organizations throughout the Bay Area. Visit http://www.ucsfhealth.org/. Follow UCSF Health on Facebook or on Twitter

Follow UCSF
ucsf.edu | Facebook.com/ucsf | YouTube.com/ucsf

Untrained beer drinkers can taste different barley genotypes

WASHINGTON STATE UNIVERSITY

Research News

IMAGE: A SET OF EXPERIMENTAL SMASH BEERS BREWED WITH A SINGLE MALT OF BARLEY AND SINGLE HOPS-- THE HOPS WERE ALL THE SAME, BUT EACH BEER CONTAINED A DIFFERENT TYPE OF... view more 

CREDIT: WSU

PULLMAN, Wash. - When it comes to craft beer, the flavor doesn't have to be all in the hops. As a panel of amateur beer tasters at Washington State University recently demonstrated, malted barley, the number one ingredient in beer besides water, can have a range of desirable flavors too.

Researchers recruited a panel of about 100 craft beer drinkers to taste some so-called SMaSH beers--those brewed with a single barley malt and single hop. All the beers contained the same hop variety, called Tahoma, but each had a malt from a different barley genotype, or genetic makeup. Trained tasters can distinguish these easily, but even the untrained panel could taste the difference among five different barley varieties--and definitely favored some more than others.

"We found that the untrained panelists could differentiate among the barley breeding lines in the beer," said Evan Craine, a WSU doctoral student and first author on the study in the Journal of Food Science. "They did a good job of selecting attributes that revealed distinctive profiles for each of the beers."

The panel generally preferred the four barley breeding lines developed at WSU over the control, known as Copeland, a high-quality malting barley widely grown in Washington state. The panelists were able to easily identify the flavor profiles of the beers, such as one with a "fruity and sweet aromatic" flavor and another with a "citrus" profile made with a barley called Palmer, a variety recently released by WSU for commercial use.

While the untrained panel could distinguish flavors from brewed beers, they were not as adept at tasting the differences among "hot steep" samples which are made by combining hot water and ground barley malt before filtering. This creates a sweet liquid--similar to that made by brewers before yeast is added to create alcohol.

The researchers had hoped amateur beer tasters could distinguish flavor differences in the hot steep as it would shorten the testing process for new barley varieties. Corresponding author Kevin Murphy was not ready to give up on the method.

"Hot steep malt still shows a lot of promise," said Murphy, a WSU associate professor of crop and soil sciences. "The next step would be testing it with a trained panel to see if they can distinguish barley varieties. Ideally, we would just set it out to consumers because hot steep malting is great outreach. It gets people involved. They love tasting and talking about it."

While U.S. craft beer drinkers are known for their love of hop-heavy India Pale Ales, the results of this study add evidence that barley malts might be another good way to develop new beers.

For this study, Craine and Murphy worked with Scott Fisk, a faculty research assistant at Oregon State University, to create the malts and brewer Aaron Hart of Moscow Brewing Company to develop the beers, using hops only to add a little bitterness to balance the sweetness from the malt. These types of beers are called "malt-forward." They can be light or strong flavored, ranging in types from lagers and pilsners to ambers and stouts. Hart called the beers developed for this study "American Pale Ales."

More variety from malt-forward beers can potentially benefit not only beer lovers but also the environment and brewers' bottom lines, said Craine.

"In terms of sustainability, hops can be pretty resource intensive, and at least around us in Pullman, we can grow barley that's just rain fed," he said. "Hops can also be really expensive. Brewers are already buying the malt, so if we can find ways to increase the flavor contribution from the malt, hopefully, they can rely less on the hops and save money."

While the hops craze is continuing, the malt-forward beers have the potential to spur the next evolution in craft brewing, said Murphy.

"Just as craft beer flavor has evolved in the last 20 years, we can expect it to continue to change over the next 20, and the new frontier will be adding different barley flavors or barley-hop combinations," he said. "I don't know how many people knew about IPAs 20 years ago, and they exploded. Brewers are very innovative, and I am very excited to see where this goes in the future."

CAPTION

WSU Ph.D. student Julianne Kellogg (left), Associate Professor Kevin Murphy and Research Associate Cedric Habiyaremye discuss the different barley genotype flavors in the beers used in the study.

CREDIT

WSU

 

New algorithm may help autonomous vehicles navigate narrow, crowded streets

CMU research could help solve last mile delivery challenges

CARNEGIE MELLON UNIVERSITY

Research News

 

IMAGE: VEHICLES ATTEMPT TO PASS EACH OTHER ON A CROWDED STREET IN PITTSBURGH, PA. RESEARCHERS AT CARNEGIE MELLON UNIVERSITY SOUGHT TO ENABLE AUTONOMOUS VEHICLES TO NAVIGATE THIS SITUATION. view more 

CREDIT: CARNEGIE MELLON UNIVERSITY

It is a scenario familiar to anyone who has driven down a crowded, narrow street. Parked cars line both sides, and there isn't enough space for vehicles traveling in both directions to pass each other. One has to duck into a gap in the parked cars or slow and pull over as far as possible for the other to squeeze by.

Drivers find a way to negotiate this, but not without close calls and frustration. Programming an autonomous vehicle (AV) to do the same -- without a human behind the wheel or knowledge of what the other driver might do -- presented a unique challenge for researchers at the Carnegie Mellon University Argo AI Center for Autonomous Vehicle Research.

"It's the unwritten rules of the road, that's pretty much what we're dealing with here," said Christoph Killing, a former visiting research scholar in the School of Computer Science's Robotics Institute and now part of the Autonomous Aerial Systems Lab at the Technical University of Munich. "It's a difficult bit. You have to learn to negotiate this scenario without knowing if the other vehicle is going to stop or go."

While at CMU, Killing teamed up with research scientist John Dolan and Ph.D. student Adam Villaflor to crack this problem. The team presented its research, "Learning To Robustly Negotiate Bi-Directional Lane Usage in High-Conflict Driving Scenarios," at the International Conference on Robotics and Automation.

The team believes their research is the first into this specific driving scenario. It requires drivers -- human or not -- to collaborate to make it past each other safely without knowing what the other is thinking. Drivers must balance aggression with cooperation. An overly aggressive driver, one that just goes without regard for other vehicles, could put itself and others at risk. An overly cooperative driver, one that always pulls over in the face of oncoming traffic, may never make it down the street.

"I have always found this to be an interesting and sometimes difficult aspect of driving in Pittsburgh," Dolan said.

Autonomous vehicles have been heralded as a potential solution to the last mile challenges of delivery and transportation. But for an AV to deliver a pizza, package or person to their destination, they have to be able to navigate tight spaces and unknown driver intentions.

The team developed a method to model different levels of driver cooperativeness -- how likely a driver was to pull over to let the other driver pass -- and used those models to train an algorithm that could assist an autonomous vehicle to safely and efficiently navigate this situation. The algorithm has only been used in simulation and not on a vehicle in the real world, but the results are promising. The team found that their algorithm performed better than current models.

Driving is full of complex scenarios like this one. As the autonomous driving researchers tackle them, they look for ways to make the algorithms and models developed for one scenario, say merging onto a highway, work for other scenarios, like changing lanes or making a left turn against traffic at an intersection.

"Extensive testing is bringing to light the last percent of touch cases," Dolan said. "We keep finding these corner cases and keep coming up with ways to handle them."

###

A machine learning breakthrough: using satellite images to improve human lives

Berkeley-based project could support action worldwide on climate, health and poverty

UNIVERSITY OF CALIFORNIA - BERKELEY

Research News

IMAGE: DEEP STREAMS OF DATA FROM EARTH-IMAGING SATELLITES ARRIVE IN DATABASES EVERY DAY, BUT ADVANCED TECHNOLOGY AND EXPERTISE ARE REQUIRED TO ACCESS AND ANALYZE THE DATA. NOW A NEW SYSTEM, DEVELOPED... view more 

CREDIT: (PHOTO BY NASA VIA PXFUEL)

Berkeley -- More than 700 imaging satellites are orbiting the earth, and every day they beam vast oceans of information -- including data that reflects climate change, health and poverty -- to databases on the ground. There's just one problem: While the geospatial data could help researchers and policymakers address critical challenges, only those with considerable wealth and expertise can access it.

Now, a team based at the University of California, Berkeley, has devised a machine learning system to tap the problem-solving potential of satellite imaging, using low-cost, easy-to-use technology that could bring access and analytical power to researchers and governments worldwide. The study, "A generalizable and accessible approach to machine learning with global satellite imagery," was published today (Tuesday, July 20) in the journal Nature Communications.

"Satellite images contain an incredible amount of data about the world, but the trick is how to translate the data into usable insights without having a human comb through every single image," said co-author Esther Rolf, a final-year Ph.D. student in computer science. "We designed our system for accessibility, so that one person should be able to run it on a laptop, without specialized training, to address their local problems."

"We're entering a regime in which our actions are having truly global impact," said co-author Solomon Hsiang, director of the Global Policy Lab at the Goldman School of Public Policy. "Things are moving faster than they've ever moved in the past. We're changing resource allocations faster than ever. We're transforming the planet. That requires a more responsive management system that is able to see these things happen, so that we can respond in a timely, effective way."

The project was a collaboration between the Global Policy Lab, which Hsiang directs, and Benjamin Recht's research team in the department of Electrical Engineering and Computer Sciences. Other co-authors are Berkeley Ph.D. graduates Tamma Carleton, now at University of California, Santa Barbara; Jonathan Proctor, now at Harvard's Center for the Environment and Data Science Initiative; Ian Bolliger, now at the Rhodium Group; and Vaishaal Shankar, now at Amazon; and Berkeley Ph.D. student Miyabi Ishihara.

All of them were at Berkeley when the project began. Their collaboration has been remarkable for bringing together disciplines that often look at the world in different ways and speak different languages: computer science, environmental and climate science, statistics, economics and public policy.

But they have been guided by a common interest in creating an open access tool that democratizes the power of technology, making it usable even by communities and countries that lack resources and advanced technical skill. "It's like Ford's Model T, but with machine learning and satellites," Hsiang said. "It's cheap enough that everyone can now access this new technology."

--MOSAIKS: Improving lives, protecting the planet--

The system that emerged from the Berkeley-based research is called MOSAIKS, short for Multi-Task Observation using Satellite Imagery & Kitchen Sinks. It ultimately could have the power to analyze hundreds of variables drawn from satellite data -- from soil and water conditions to housing, health and poverty -- at a global scale.

The research paper details how MOSAIKS was able to replicate with reasonable accuracy reports prepared at great cost by the U.S. Census Bureau. It also has enormous potential in addressing development challenges in low-income countries and to help scientists and policymakers understand big-picture environmental change.

"Climate change is diffuse and difficult to see at any one location, but when you step back and look at the broad scale, you really see what is going on around the planet," said Hsiang, who also serves as co-director of the multi-institution Climate Impact Lab.

For example, he said, the satellite data could give researchers deep new insights into expansive rangeland areas such as the Great Plains in the U.S. and the Sahel in Africa, or into areas such as Greenland or Antarctica that may be shedding icebergs as temperatures rise.

"These areas are so large, and to have people sitting there and looking at pictures and counting icebergs is really inefficient," Hsiang explained. But with MOSAIKS, he said, "you could automate that and track whether these glaciers are actually disintegrating faster, or whether this has been happening all along."

For a government in the developing world, the technology could help guide even routine decisions, such as where to build roads.

"A government wants to build roads where the most people are and the most economic activity is," Hsiang said. "You might want to know which community is underserved, or the condition of existing infrastructure in a community. But often it's very difficult to get that information."

--The challenge: Organizing trillions of bytes of raw satellite data--

The growing fleet of imaging satellites beam data back to Earth 24/7 -- some 80 terabytes every day, according to the research, a number certain to grow in coming years.

But often, imaging satellites are built to capture information on narrow topics -- supplies of fresh water, for example, or the condition of agricultural soils. And the data doesn't arrive as neat, orderly images, like a snapshots from a photo shop. It's raw data, a mass of binary information. Researchers who access the data have to know what they're looking for.

Merely storing so many terabytes of data requires a huge investment. Distilling the layers of data embedded in the images requires additional computing power and advanced human expertise to tease out strands of information that are coherent and useful to other researchers, policymakers or funding agencies.

Inevitably, exploiting satellite images is largely limited to scholars or agencies in wealthy nations, Rolf and Hsiang said.

"If you're an elite professor, you can get someone to build your satellite for you," said Hsiang. "But there's no way that a conservation agency in Kenya is going to be able to access the technology and the experts to do this work.

"We wanted to find a way to empower them. We decided to come up with a Swiss Army Knife -- a practical tool that everyone can access."

--Like Google for satellite imagery, sort of--

Especially in low-income countries, one dimension of poverty is a poverty of data. But even communities in the U.S. and other developed countries usually don't have ready access to geospatial data in a convenient, usable format for addressing local challenges.

CAPTION

In the Indian state of Andhra Pradesh, a satellite image shows hundreds of green aquaculture ponds where local farmers grow fish and shrimp. Geospatial imaging holds enormous potential for developing nations to address challenges related to agriculture, poverty, health and human migration, scholars at UC Berkeley say. But until now, the technology and expertise needed to efficiently access and analyze satellite data usually has been limited to developed countries.

CREDIT

(NASA Earth Observatory images by Joshua Stevens, using Landsat data from the U.S. Geological Survey.)

Machine learning opens the door to solutions.

In a general sense, machine learning refers to computer systems that use algorithms and statistical modeling to learn on their own, without step-by-step human intervention. What the new research describes is a system that can assemble data delivered by many satellites and organize it in ways that are accessible and useful.

There are precedents for such systems: Google Earth Engine and Microsoft's Planetary Computer are both platforms for accessing and analyzing global geospatial data, with a focus on conservation. But, Rolf said, even with these technologies, considerable expertise is often required to convert the data into new insights.

The goal of MOSAIKS is not to develop more complex machine learning systems, Rolf said. Rather, its innovation is in making satellite data widely useable for addressing global challenges. The team did this by making the algorithms radically simpler and more efficient.

MOSAIKS starts with learning to recognize minuscule patterns in the images -- Hsiang compares it to a game of Scrabble, in which the algorithm learns to recognize each letter. In this case, however, the tiles are minuscule pieces of satellite image, 3 pixels by 3 pixels.

But MOSAIKS doesn't conclude "this is a tree" or "this is pavement." Instead, it recognizes patterns and groups them together, said Proctor. It learns to recognize similar patterns in different parts of the world.

When thousands of terabytes from hundreds of sources are analyzed and organized, researchers can choose a village or a country or a region and draw out organized data that can touch on themes as varied as soil moisture, health conditions, human migration and home values.

In a sense, Hsiang said, MOSAIKS could do for satellite databases what Google in the early days did for the Internet: map the data, make it accessible and user-friendly at low cost, and perhaps make it searchable. But Rolf, a machine learning scholar based in the Berkeley Electrical Engineering and Computer Sciences department, said the Google comparison goes only so far.

MOSAIKS "is about translating an unwieldy amount of data into usable information," she explained. "Maybe a better analogy would be that the system takes very dense information -- say, a very large article -- and produces a summary."

--Creating a living atlas of global data--

Both Hsiang and Rolf see the potential for MOSAIKS to evolve in powerful and elegant directions.

Hsiang imagines the data being collected into computer-based, continually evolving atlases. Turn to any given "page," and a user could access broad, deep data about conditions in a country or a region.

Rolf envisions a system that can take the stream of data from humanity's fleet of imaging satellites and remote sensors and transform it into a flowing, real-time portrait of Earth and its inhabitants, continually in a state of change. We could see the past and the present, and so discern emerging challenges and address them.

"We've sent so much stuff to space," Hsiang says. "It's an amazing achievement. But we can get a lot more bang for our buck for all of this data that we're already pulling down. Let's let the world use it in a useful way. Let's use it for good."

###