Research method
Logistic Regression
15 studies
Logistic regression models the log-odds of a binary outcome — fully vaccinated or not, C. difficile infection or not, lifestyle-adherent or not — as a linear function of covariates. The headline number is usually an odds ratio or adjusted odds ratio (AOR): AOR 3.10 means about three-fold higher odds after the model's other covariates are held fixed. The same binary-outcome machinery also sits underneath many diagnostic risk scores, where the useful number may be an AUC or a positive predictive value rather than any one AOR.
Epidemiologists reach for logistic models when the outcome is yes/no at a defined window rather than time-to-event. It answers 'which factors are associated with higher odds of this event, after adjustment?' Its main limitation is design, not software: an AOR from a DHS or EHR analysis is not a randomised treatment effect, and an AOR from a trial is only as durable as the follow-up and as trustworthy as a self-reported endpoint.
Studies
15
Findings
5
7 supporting · 0 challenging · 0 qualifying citations
Open tensions
2
Currently
What we know
- In Senegal's 2010–2011 DHS, complete immunization among 2,199 children aged 12–23 months was 62.8%, below national (>80%) and international (>90%) goals. Four antenatal visits were associated with AOR 3.10 for complete immunization (card or maternal recall). That is an adjusted association in survey data, not a trial of an outreach programme, and recall can misclassify doses.
- Linked Health and Retirement Study–Medicare analyses found CDI rates of 282.9 versus 197.1 per 100,000 person-years with versus without depression, and adjusted odds of CDI 36% higher with major depression; widowhood and living alone were also associated. The model does not show that treating depression would lower CDI, and residual confounding by healthcare exposure remains possible.
- A randomised AOR can still fade. New Zealand adults with coronary heart disease assigned to Text4Heart had higher lifestyle-adherence odds at 3 months (AOR 2.55) that were no longer significant at 6 months. Medication-adherence scores improved; LDL differences were only borderline. The trial was modest, relied on self-report for the primary behaviour outcome, and does not prove fewer hard cardiac events.
- Binary classification papers in this set use related logistic/ROC machinery for detection rather than for a single exposure AOR. Among 50,780 English primary-care CA125 tests, ovarian-cancer incidence was 0.9%; at ≥35 U/ml, sensitivity was 77%, specificity 93.8%, AUC 0.92, and PPV only 10.1%. In Amsterdam, diabetes prevalence was 25.6%, 12.7% and 6.8% across three ethnic groups, with risk-score AUCs 0.74, 0.80 and 0.78 and numbers needed to screen of 3, 5 and 7.
- Logistic-style binary contrasts also describe coverage and access gaps without naming a causal lever. Among 1,846 New Zealand women with first primary breast cancer, screen-detection was 62.7% in NZ European versus 49.2% in Māori cancers. In a national sample of 18,215 middle-aged and older Chinese adults, only 6.5% had recent Internet access while 83% owned a mobile phone; Internet access tracked neighbourhood amenities and health disparities more than phones did.
Largest unresolved question
The same odds-ratio language is asked to do two different jobs. Senegal's AOR 3.10 and the CDI 36% excess odds are observational associations after covariate adjustment. Text4Heart's AOR 2.55 is a randomised contrast that itself was gone by 6 months. Reading every AOR as if it were that trial overclaims the survey papers and understates how fragile a trial AOR can be once follow-up continues.
Procedure
How it works
The main stages of the method. Specific protocols and analysis choices vary between studies.
Code a binary outcome at a defined window
The outcome is yes/no at a fixed window — fully vaccinated or not, infection or not, screen-detected or not — not time-to-event. Diagnostic papers sometimes use the same machinery for a cutoff rather than for an exposure.
Specify covariates on the log-odds scale
Log-odds of the event is written as a linear function of covariates. Which confounders enter, and how they are coded, is a scientific choice, not a software default.
Fit the model and report an odds ratio
Coefficients exponentiate to odds ratios. AOR 3.10 means about three-fold higher odds after the model's other covariates are held fixed.
Read the number inside its design
An AOR from a survey or EHR analysis is an adjusted association; an AOR from a randomised programme is a contrast that can fade when follow-up continues; an AUC or predictive value from a diagnostic score is an operating point, not an exposure odds ratio.
Do not treat every AOR as a trial effect
Logistic regression answers which factors are associated with higher odds of this event after adjustment. Design, not the fitting routine, decides whether that number can be read as a treatment effect.
Common misconceptions
An AOR of 3.10 means four antenatal visits cause complete childhood immunization.
It is an adjusted association in DHS data for 2,199 Senegalese children. The analysis cannot identify the causal effect of a specific outreach programme, and recall-based vaccination histories can misclassify doses.
If a trial reports AOR 2.55, the behaviour change is locked in.
Text4Heart's lifestyle-adherence contrast was significant at 3 months and not at 6 months. An AOR is tied to a time point and to a self-reported endpoint; it is not a hard-event result.
AUC 0.92, or similar AUCs across groups, means the test is about equally useful for the next patient.
CA125's AUC 0.92 sat next to a PPV of 10.1% because incidence was 0.9%. The diabetes score's AUCs (0.74–0.80) sat next to NNS of 3, 5 and 7 because prevalence differed several-fold. Usefulness is not the AUC alone.
Related
Claim ledger
What the evidence shows
Drawn from 15 studies in this library. Mix labels say which citation roles are present; they are not a strength score. Supports means evidence for a finding; Challenges means evidence against a stated position; Qualifies marks scope.
In Senegal's 2010–2011 DHS, complete immunization among 2,199 children aged 12–23 months was 62.8%, below national (>80%) and international (>90%) goals. Four antenatal visits were associated with AOR 3.10 for complete immunization (card or maternal recall). That is an adjusted association in survey data, not a trial of an outreach programme, and recall can misclassify doses.
Linked Health and Retirement Study–Medicare analyses found CDI rates of 282.9 versus 197.1 per 100,000 person-years with versus without depression, and adjusted odds of CDI 36% higher with major depression; widowhood and living alone were also associated. The model does not show that treating depression would lower CDI, and residual confounding by healthcare exposure remains possible.
A randomised AOR can still fade. New Zealand adults with coronary heart disease assigned to Text4Heart had higher lifestyle-adherence odds at 3 months (AOR 2.55) that were no longer significant at 6 months. Medication-adherence scores improved; LDL differences were only borderline. The trial was modest, relied on self-report for the primary behaviour outcome, and does not prove fewer hard cardiac events.
Binary classification papers in this set use related logistic/ROC machinery for detection rather than for a single exposure AOR. Among 50,780 English primary-care CA125 tests, ovarian-cancer incidence was 0.9%; at ≥35 U/ml, sensitivity was 77%, specificity 93.8%, AUC 0.92, and PPV only 10.1%. In Amsterdam, diabetes prevalence was 25.6%, 12.7% and 6.8% across three ethnic groups, with risk-score AUCs 0.74, 0.80 and 0.78 and numbers needed to screen of 3, 5 and 7.
Study Role Design N Population Outcome How well does CA125 find ovarian cancer in GP care? Supports CohortPopulation-based primary-care EHR cohort N=50780 · Women with CA125 tested in English primary care Women having CA125 measured in English general practice PPV, sensitivity, specificity, and AUC of CA125 ≥35 U/ml for ovarian cancer Diabetes prevalence and risk-score accuracy Supports Cross-sectionalSUNSET Amsterdam population sample; ethnicity-stratified screening criteria N=1434 · 339 Hindustani Surinamese, 605 African Surinamese, 490 Dutch Amsterdam adults aged 35–60 (Hindustani Surinamese, African Surinamese, Dutch) Diabetes prevalence and risk-score AUC / numbers-needed-to-screen by ethnicity Logistic-style binary contrasts also describe coverage and access gaps without naming a causal lever. Among 1,846 New Zealand women with first primary breast cancer, screen-detection was 62.7% in NZ European versus 49.2% in Māori cancers. In a national sample of 18,215 middle-aged and older Chinese adults, only 6.5% had recent Internet access while 83% owned a mobile phone; Internet access tracked neighbourhood amenities and health disparities more than phones did.
Study Role Design N Population Outcome Screening and breast cancer equity in NZ Supports CohortNew Zealand screening-age women with first primary breast cancer; screen-detected vs symptomatic N=1846 · 1,548 invasive and 298 in-situ; 1,064 (57.6%) screen-detected Screening-age women in New Zealand with newly diagnosed first primary breast cancer Screen detection rates and cancer-specific survival by ethnicity Is China’s digital divide tied to health gaps? Supports Cross-sectionalCHARLS wave relating Internet and mobile-phone access to socioeconomic position and health N=18215 Nationally representative middle-aged and older Chinese adults (45+) Health outcomes associated with Internet access and mobile-phone ownership
Debates
Tensions and limits
Some items are genuine disagreements on the same question. Others mark different assays, populations, or outcomes.
The same odds-ratio language is asked to do two different jobs. Senegal's AOR 3.10 and the CDI 36% excess odds are observational associations after covariate adjustment. Text4Heart's AOR 2.55 is a randomised contrast that itself was gone by 6 months. Reading every AOR as if it were that trial overclaims the survey papers and understates how fragile a trial AOR can be once follow-up continues.
The same odds-ratio language is asked to do two different jobs. Senegal's AOR 3.10 and the CDI 36% excess odds are observational associations after covariate adjustment. Text4Heart's AOR 2.55 is a randomised contrast that itself was gone by 6 months. Reading every AOR as if it were that trial overclaims the survey papers and understates how fragile a trial AOR can be once follow-up continues.
- What predicts full childhood immunization in Senegal
- Depression linked to C. difficile infection risk
- Do heart texts improve lifestyle adherence?
Study Role Design N Population Outcome What predicts full childhood immunization in Senegal Supports Cross-sectionalSecondary analysis of Senegal 2010–2011 DHS N=2199 · Children aged 12–23 months Senegalese children aged 12–23 months in the Demographic and Health Survey Complete immunization coverage and maternal/household predictors Depression linked to C. difficile infection risk Supports CohortLinked HRS–Medicare analyses plus complementary clinical cohorts N=16781 · HRS–CMS linked Study 1 sample size; complementary clinical cohorts also reported Older US adults in the Health and Retirement Study linked to Medicare claims Clostridium difficile infection rates/odds by depression and antidepressant exposure Do heart texts improve lifestyle adherence? Supports RCTText4Heart 24-week mHealth cardiac rehabilitation plus usual care vs usual care alone N=123 · 61 intervention, 62 control New Zealand adults with coronary heart disease Adherence to recommended lifestyle behaviours at 3 and 6 months
PaperFren reads this as a limit on how far one study travels — different assays, populations, or outcomes — not a forced fight between papers.
Diagnostic papers optimise a cutoff, not an exposure odds ratio. CA125's PPV of 10.1% at a guideline threshold is a rare-disease problem; the diabetes score's NNS of 3–7 is a screening-workload problem that tracks prevalence, not AUC. Those operating-point numbers are not interchangeable with an immunization AOR of 3.10.
Diagnostic papers optimise a cutoff, not an exposure odds ratio. CA125's PPV of 10.1% at a guideline threshold is a rare-disease problem; the diabetes score's NNS of 3–7 is a screening-workload problem that tracks prevalence, not AUC. Those operating-point numbers are not interchangeable with an immunization AOR of 3.10.
- How well does CA125 find ovarian cancer in GP care?
- Diabetes prevalence and risk-score accuracy
- What predicts full childhood immunization in Senegal
Study Role Design N Population Outcome How well does CA125 find ovarian cancer in GP care? Supports CohortPopulation-based primary-care EHR cohort N=50780 · Women with CA125 tested in English primary care Women having CA125 measured in English general practice PPV, sensitivity, specificity, and AUC of CA125 ≥35 U/ml for ovarian cancer Diabetes prevalence and risk-score accuracy Supports Cross-sectionalSUNSET Amsterdam population sample; ethnicity-stratified screening criteria N=1434 · 339 Hindustani Surinamese, 605 African Surinamese, 490 Dutch Amsterdam adults aged 35–60 (Hindustani Surinamese, African Surinamese, Dutch) Diabetes prevalence and risk-score AUC / numbers-needed-to-screen by ethnicity What predicts full childhood immunization in Senegal Supports Cross-sectionalSecondary analysis of Senegal 2010–2011 DHS N=2199 · Children aged 12–23 months Senegalese children aged 12–23 months in the Demographic and Health Survey Complete immunization coverage and maternal/household predictors
PaperFren reads this as a limit on how far one study travels — different assays, populations, or outcomes — not a forced fight between papers.
Timeline
How understanding moved
Study years are when the paper was published. Evidence edits are dated changes to this page's claims. Explanations are when PaperFren added a Discovery — not a claim that the science happened that day.
2026
Concept page published
Logistic Regression
Change log
What changed
Dated edits to this page's evidence: studies added or removed from a claim, claims added or withdrawn, and new explanations tagged here. Coverage edits are distinct from reviewed research changes and corrections. Dates record PaperFren updates; study publication dates appear in the timeline.
- Coverage updated · Concept page published
Papers
15 studies in this library bear on Logistic Regression, ordered by citations.
- COVID-19 death risk and comorbidities
In 31,461 US adults with COVID-19, older age, male sex, Black race, and several comorbidities including heart disease predicted higher mortality.
- Does higher BMI causally raise heart disease risk?
Genetic instruments for BMI suggest a causal increase in ischemic heart disease risk larger than simple observational estimates.
- Do heart texts improve lifestyle adherence?
A text-and-internet cardiac rehab program improved short-term lifestyle adherence after CHD, but the benefit faded by 6 months.
- StEP pain-subtype assessment
A structured interview-plus-exam tool (StEP) was developed and validated to separate neuropathic/radicular from non-neuropathic axial low-back pain.
- Parents and school HPV uptake
In BC’s free school HPV program, about 65% of surveyed parents reported first-dose receipt—lower than concurrent hepatitis B and meningitis C consent rates.
- COVID vaccine acceptance in Libya
Willingness to take a COVID-19 vaccine rose with assumed efficacy—about 80% if efficacy was ≥90%, but only about 41% if 50%.
- Is China’s digital divide tied to health gaps?
In older Chinese adults, Internet use was rare and SES-linked, while mobile phones were common—shaping how digital tools might reduce health disparities.
- Does a step-log app keep people tracking activity?
In 10,000 Steps members, a smartphone iStepLog app preserved daily logging and high step counts versus matched controls whose logging fell.
Show 7 more studiesShow fewer studies
- How well does CA125 find ovarian cancer in GP care?
In UK primary care, CA125 ≥35 U/ml had high NPV but only about 10% PPV for ovarian cancer, and elevated values also flagged other cancers.
- COVID-19 vaccine intention in BC
About 80% of surveyed British Columbians said they were somewhat or very likely to take a recommended COVID-19 vaccine.
- Seasonal vs pandemic flu vaccine coverage
In France, seasonal influenza coverage stayed near 21% while pandemic H1N1 coverage was only about 11%.
- Diabetes prevalence and risk-score accuracy
Hindustani Surinamese had ~26% diabetes prevalence; a clinical risk score showed moderate-to-good AUCs (~0.74–0.80) across ethnic groups.
- What predicts full childhood immunization in Senegal
Only 62.8% of Senegalese children 12–23 months were fully immunized; maternal education, wealth, and antenatal care strongly predicted coverage.
- Depression linked to C. difficile infection risk
Adults with major depression had higher CDI rates, and specific antidepressants plus living alone/widowhood also associated with infection risk.
- Screening and breast cancer equity in NZ
Screen-detected cancers were less common in Māori women, and crude survival was worse, but survival gaps narrowed among screen-detected cases.
Compare studies
Select 2–10 studies. Design and N are labels, not a ranking.
Nothing selected yet.
Questions
What is still open
The same odds-ratio language is asked to do two different jobs. Senegal's AOR 3.10 and the CDI 36% excess odds are observational associations after covariate adjustment. Text4Heart's AOR 2.55 is a randomised contrast that itself was gone by 6 months. Reading every AOR as if it were that trial overclaims the survey papers and understates how fragile a trial AOR can be once follow-up continues.
Diagnostic papers optimise a cutoff, not an exposure odds ratio. CA125's PPV of 10.1% at a guideline threshold is a rare-disease problem; the diabetes score's NNS of 3–7 is a screening-workload problem that tracks prevalence, not AUC. Those operating-point numbers are not interchangeable with an immunization AOR of 3.10.
Ask PaperFren about Logistic Regression
Study this conceptflashcards and short-answer questions
Senegal's complete-immunization coverage was 62.8% and four antenatal visits had AOR 3.10. What question does that AOR answer, and what question would require a different study?
It answers whether, after adjustment in the 2010–2011 DHS sample of 2,199 children, four antenatal visits were associated with higher odds of complete immunization. It does not answer whether a named outreach programme would raise coverage to the >80% national or >90% international goals, and recall-based histories can still misclassify doses.
Why is Text4Heart's lifestyle AOR of 2.55 not interchangeable with the CDI paper's 36% higher odds of infection among people with major depression?
Text4Heart randomised a 24-week mHealth programme; the AOR is a randomised contrast at 3 months that was gone by 6 months. The CDI figure is an observational adjusted association (rates 282.9 vs 197.1 per 100,000 person-years) that cannot show treating depression would prevent CDI.
A colleague says CA125 'works' because AUC is 0.92. Which other number from the same cutoff should they quote to a GP, and why?
PPV 10.1% at ≥35 U/ml, with incidence 0.9%, sensitivity 77% and specificity 93.8%. Most positive tests were not ovarian cancer. AUC ranks cases versus non-cases across thresholds; PPV is what a positive result means at the cutoff used in practice.
The Amsterdam diabetes risk score has AUCs 0.74, 0.80 and 0.78. Why might screening burden still differ sharply across the three groups?
Prevalence was 25.6%, 12.7% and 6.8%, so numbers needed to screen were 3, 5 and 7. Similar discrimination does not mean similar tests per case found.