Risks for impaired post-stroke cognitive function

In a printed posted to the medRxiv preprint archive this month, I found a chart review of patients with stroke to determine factors (other than the stroke itself) which could predict cognitive impairment following the stroke admission to hospital.
The paper used patients admitted to Massachusetts General Hospital as their cases for which to form a risk factor formula, and then validated the formula against cases in a sister Boston hospital, Brighamm and Women’s Hospital.
The results are quoted as a table below:

The results found were that older age was the biggest overall risk factor (weighted for example as 4 for > 65, 8 for > 85), followed by those weighted 2: insurance of Medicare (a proxy for age over 65), mobility problems, delirium, peripheral vascular disease, and depression. Other factors were rated only at a weight of 1: these included Medicaid(a proxy for low income), prior falls, Parkinson’s, kidney disease, weight loss, and nursing home placement from the acute hospital stay.
Indeed, it seems that the factors found more significant, such as age, gait status, depression, delirium, vascular disease, are actually all also known from other, prospective studies to be risk factors for dementia in the general population, even in those without any stroke.
==================================================================================================
ABSTRACT
Predicting post-stroke cognitive impairment using electronic health record data
Jeffrey M. Ashburner, PhD, MPH1,2; Yuchiao Chang, PhD1,2; Bianca Porneala, MS1; Sanjula D. Singh, MD, PhD3; Nirupama Yechoor, MD, MSc3; Jonathan M. Rosand, MD, MSc3; Daniel E. Singer, MD1,2; Christopher D. Anderson, MD, MMSc4; Steven J. Atlas, MD, MPH
medRxiv preprint: https://doi.org/10.1101/2024.02.02.24302240
Importance: Secondary prevention interventions to reduce post-stroke cognitive impairment (PSCI) can be aided by the early identification of high-risk individuals who would benefit from risk factor modification.
Objective: To develop and evaluate a predictive model to identify patients at increased risk of PSCI over 5 years using data easily accessible from electronic health records.
Design: Cohort study with patients enrolled between 2003-2016 with follow-up through 2022. Setting: Primary care practices affiliated with two academic medical centers.
Participants: Individuals 45 years or older, without prior stroke or prevalent cognitive impairment, with primary care visits and an incident ischemic stroke between 2003-2016 (development/internal validation cohort) or 2010-2022 (external validation cohort).
Exposures: Predictors of PSCI were ascertained from the electronic health record.
Main Outcome: The outcome was incident dementia/cognitive impairment within 5 years and beginning 3 months following stroke, ascertained using ICD-9/10 codes. For model variable selection, we considered potential predictors of PSCI and constructed 400 bootstrap samples with two-thirds of the model derivation sample. We ran 10-fold cross-validated Cox proportional hazards models using a least absolute shrinkage and selection operator (LASSO) penalty. Variables selected in >25% of samples were included. Results: The analysis included 332 incident diagnoses of PSCI in the development cohort (n=3,741), and 161 and 128 incident diagnoses in the internal (n=1,925) and external (n=2,237) validation cohorts. The c-statistic for predicting PSCI was 0.731 (95% CI: 0.694-0.768) in the internal validation cohort, and 0.724 (95% CI: 0.681-0.766) in the external validation cohort. A risk score based on the beta coefficients of predictors from the development cohort stratified patients into low (0-7 points), intermediate (8-11 points), and high (12-35 points) risk groups. The hazard ratios for incident PSCI were significantly different by risk categories in internal (High, HR: 6.2, 95% CI 4.1-9.3; Intermediate, HR 2.7, 95% CI: 1.8-4.1) and external (High, HR: 6.1, 95% CI: 3.9-9.6; Intermediate, HR 2.8, 95% CI: 1.9-4.3) validation cohorts.
Conclusions and Relevance: Five-year risk of PSCI can be accurately predicted using routinely collected data. Model output can be used to risk stratify and identify individuals at increased risk for PSCI for preventive efforts.
Key Words: post-stroke cognitive impairment, risk stratification, risk prediction

On medical AI in 2023

“To study persons is to study beings who only exist in, or are partly constituted by a certain language”. -- Charles Taylor, Sources of the Self

Large language model-based AI (LLM’s) are the epitome of what can be constituted by language alone. They can easily take isolated linguistic philosophy to its absurd extreme.

Unfortunately, this means there is a disconnect between a statement that is correct in the context of its LLM and the statements we want, which appropriately address a scientific or clinical context in the world outside of the LLM. This results in AI dialog replies that currently make medical advisory AI impossible to trust.

Current LLM base AI’s are masters of what Harry G. Frankfurt called “bullshit.” Until AI can distinguish and eliminate fictional or obsolete diagnoses and treatments within its model, separating them from those which help the patient, it cannot be trusted with any unsupervised role in patient care.

Dammed lies and statistics

Picture of Itajai, Brazil

Mark Twain wrote: “There are three kinds of lies: lies, damned lies, and statistics.”

Statistics continue to mislead the unwary in the current milieu of rapid online publication of articles and preprints. A case in point is the study published last month in the fast-tracking online medical journal Cureus. This study, entitled Regular Use of Ivermectin as Prophylaxis for COVID-19 Led Up to a 92% Reduction in COVID-19 Mortality Rate in a Dose-Response Manner: Results of a Prospective Observational Study of a Strictly Controlled Population of 88,012 Subjects, purported to show that prolonged, regular use of ivermectin lessened the incidence and mortality of COVID-19 as measured by total dose taken in the period before infection occurred.

Note that though the study title states it was prospective, planning a study in advance is not sufficient for a study to be prospective. Because subjects were only placed into their treatment groups after the study outcomes were known, this was a retrospective study.

The problem lies in the study's design: total dosage of ivermectin over the entire period before COVID infection was the regression variable, versus mortality in those infected. However, because intention-to-treat analysis was thus bypassed, the study was actually measuring the mortality in the group which took ivermectin for many months and either never got COVID-19 or got COVID-19 only after that period, versus the mortality in those who either never took ivermectin or who took ivermectin for only a short time. This means that during the early months of this study, almost all of the patients who developed COVID-19 were placed in the non-ivermectin or the "irregular user" group even if they would have been placed in the ivermectin group had they NOT come down with COVID (and thus have more time to take doses of ivermectin, as many ultimately did). In other words, the fact that some who got COVID-19 before taking ivermectin might well have taken ivermectin later in any given study period had they not gotten COVID-19 first was ignored.

Was this study "strictly controlled"? Yes, in a manner that biases it beyond repair, unless they had ALSO given intent-to-treat data! Only the authors know why their design was designed the way it was.

----------------------------------------------------------------------------------

REFERENCE: Kerr L, Baldi F, Lobo R, et al. (August 31, 2022) Regular Use of Ivermectin as Prophylaxis for COVID-19 Led Up to a 92% Reduction in COVID-19 Mortality Rate in a Dose-Response Manner: Results of a Prospective Observational Study of a Strictly Controlled Population of 88,012 Subjects. Cureus 14(8): e28624.

DOI: 10.7759/cureus.28624

---------------------------------------------------------------------------------

Here is a computer simulation (written in Julia) which demonstates the study bias mentioned above. In the program, the treatment is set to have no actual effect on the infection rate. The highly significant results are purely from the bias of excluding those who become infected from later entering a treatment arm.

using HypothesisTests

@enum TreatmentClass Untreated Irregular Excluded Regular

mutable struct Subject
    cum_dose::Float64
    treatment_class::TreatmentClass
    had_covid::Bool
    last_dose_day::Int
end

function update!(subjects::Vector{Subject}, day, pcovid = 0.001, pstart = 0.0075, pdosing = 0.25, dosage = 35)
    for subj in subjects
        if subj.had_covid
            continue
        elseif rand() < pcovid
            subj.had_covid = true
        elseif (subj.cum_dose > 0 && rand() <= pdosing  && (day > subj.last_dose_day + 14 || day == subj.last_dose_day + 1)) ||
           (subj.cum_dose == 0 && rand() < pstart)
            subj.cum_dose += dosage
            subj.last_dose_day = day
            subj.treatment_class =
               subj.cum_dose == 0 ? Untreated : subj.cum_dose >= 180 ? Regular : subj.cum_dose <= 60 ? Irregular : Excluded
        end
    end
end

function run_study(N = 10_000, duration = 150)
    population = [Subject(0.0, Untreated, false, 0) for _ in 1:N]
    unt, unt_covid, irr, irr_covid, reg, reg_covid, excluded = 0, 0, 0, 0, 0, 0, 0
    println("Population size $N, daily infection risk 0.1%")
    for day in 1:duration
        update!(population, day)
        if day % 30 == 0
            println("\nDay $day:")
            unt = count(s -> s.treatment_class == Untreated, population)
            unt_covid = count(s -> (s.treatment_class == Untreated) && s.had_covid, population)
            println("Untreated: N = $unt, with infection = $unt_covid")
            irr = count(s -> s.treatment_class == Irregular, population)
            irr_covid = count(s -> (s.treatment_class == Irregular) && s.had_covid, population)
            println("Irregular Use: N = $irr, with infection = $irr_covid")
            reg = count(s -> s.treatment_class == Regular, population)
            reg_covid = count(s -> (s.treatment_class == Regular) && s.had_covid, population)
            println("Regular Use: N = $reg, with infection = $reg_covid")
            exc = count(s -> (s.treatment_class == Excluded) && s.had_covid, population)
            println("Excluded: N = $exc")
        end
        if day == 75
            println("\nAt midpoint, Infection case percentages are:")
            println("  Untreated : ", Float16(100 * unt_covid / unt))
            println("  Irregulars: ", Float16(100 * irr_covid / irr))
            println("  Regulars  : ", Float16(100 * reg_covid / reg))
        end
    end
    println("\nAt study end, Infection case percentages are:")
    println("  Untreated : ", Float16(100 * unt_covid / unt), " of group size of $unt")
    println("  Irregulars: ", Float16(100 * irr_covid / irr), " of group size of $irr")
    println("  Regulars  : ", Float16(100 * reg_covid / reg), " of group size of $reg")
    untreated = [s.had_covid for s in population if s.treatment_class == Untreated]
    irregular = [s.had_covid for s in population if s.treatment_class == Irregular]
    regular = [s.had_covid for s in population if s.treatment_class == Regular]
    excluded = [s.had_covid for s in population if s.treatment_class == Excluded]
    println("\n\n   Final statistics:\n")
    @show KruskalWallisTest(untreated, irregular, regular, excluded)
end

run_study()

Treating late-presenting strokes from large vessel occlusion

Only a small minority of large vessel strokes which might be eligible for TPA thrombolysis of clot extraction removal of blockage are so treated. WOne reason is that such strokes often have a stuttering course which causes them to present to the ER only well after they have begun, or to present with changes on CT from previously begun tissue infaction even when deficits are small.

Unfortunately, many such persons with stroke worsen in the hospital. The study below offers hope for late treatment of at least some of these patients to reduce their stroke burden.

=============================================================

ABSTRACT

Identifying Severe Stroke Patients Likely to Benefit From Thrombectomy Despite Delays of up to a Day

R. Gilberto González, Gisele Sampaio Silva, Julian He, Saloomeh Sadaghiani, Ona Wu, Aneesh B. Singhal

Scientific Reports, August 2022

doi: 10.1038/s41598-020-60933-3 0000005

Abstract

Selected patients with large vessel occlusions (LVO) can benefit from thrombectomy up to 24 hours after onset. Identifying patients who might benefit from late intervention after transfer from community hospitals to thrombectomy-capable centers would be valuable. We searched for presentation biomarkers to identify such patients. Frequent MR imaging over 2 days of 38 untreated LVO patients revealed logarithmic growth of the ischemic infarct core. In 24 patients with terminal internal carotid artery or the proximal middle cerebral artery occlusions we found that an infarct core growth rate (IGR) <4.1 ml/hr and initial infarct core volumes (ICV) <19.9 ml had accuracies >89% for identifying patients who would still have a core of <50ml 24 hours after stroke onset, a core size that should predict favorable outcomes with thrombectomy. Published reports indicate that up to half of all LVO stroke patients have an IGR<4.1 ml/hr. Other potentially useful biomarkers include the NIHSS and the perfusion measurements MTT and Tmax. We conclude that many LVO patients have a stroke physiology that is favorable for late intervention, and that there are biomarkers that can accurately identify them at early time points as suitable for transfer for intervention.

Risks for impaired post-stroke cognitive function

In a printed posted to the medRxiv preprint archive this month, I found a chart review of patients with stroke to determine factors (other t...