Recurrent interventions

Expected number of procedural interventions in the first year after radiocephalic AVF creation

Choose the model that matches the information you have: an ultrasound at 4–6 weeks, an ultrasound at about 12 weeks, or baseline characteristics before surgery. The curve shows the expected cumulative number of interventions over the first year. An intervention is any surgical or endovascular procedure on the access (for example, angioplasty), and every procedure counts, not just the first. Dashed lines mark the KDOQI 2019 intervention goals for an AV access: no more than 2 interventions to facilitate its use, and no more than 3 interventions per year to maintain its use.

Heindel P, et al. Predicting recurrent interventions after radiocephalic arteriovenous fistula creation with machine learning and the PREDICT-AVF web app. J Vasc Access. 2025;26(1):202–210.
Lok CE, et al. KDOQI Clinical Practice Guideline for Vascular Access: 2019 Update. Am J Kidney Dis. 2020;75(4 Suppl 2):S1–S164.

viewof early = ultrasoundForm("early")

resultCard(earlyCurve)

responsive((w) => interventionsPlot(earlyCurve, w), invalidation)

How the inputs are defined

Duplex ultrasound performed per the trial protocol at 4–6 weeks (early tab) or 12 weeks (late tab) after AVF creation, and read by a blinded core laboratory.

Days since AVF creation at ultrasound: Number of days after AVF creation when the ultrasound was performed.
Prior interventions: Number of interventions on the access already performed.
AVF in use for dialysis: The AVF has already been used for hemodialysis.
Flow volume: Access flow volume: the mean of three measurements in the cephalic vein 5 cm above (proximal to) the anastomosis.
Distal, mid and proximal diameter: Cephalic vein lumen diameter, measured twice and averaged at each of three forearm locations: 3 cm above the anastomosis (distal), mid-access (mid), and just below the antecubital fossa (proximal).
Outflow, juxta-anastomotic and inflow stenosis: ≥50% narrowing of the lumen anywhere along the access, recorded by location: outflow vein, juxta-anastomotic segment, or inflow artery. Tick each location where it is present.

Definitions follow the methods of Heindel et al., J Vasc Access 2025.

viewof late = ultrasoundForm("late")

resultCard(lateCurve)

responsive((w) => interventionsPlot(lateCurve, w), invalidation)

How the inputs are defined

Duplex ultrasound performed per the trial protocol at 4–6 weeks (early tab) or 12 weeks (late tab) after AVF creation, and read by a blinded core laboratory.

Days since AVF creation at ultrasound: Number of days after AVF creation when the ultrasound was performed.
Prior interventions: Number of interventions on the access already performed.
AVF in use for dialysis: The AVF has already been used for hemodialysis.
Flow volume: Access flow volume: the mean of three measurements in the cephalic vein 5 cm above (proximal to) the anastomosis.
Distal, mid and proximal diameter: Cephalic vein lumen diameter, measured twice and averaged at each of three forearm locations: 3 cm above the anastomosis (distal), mid-access (mid), and just below the antecubital fossa (proximal).
Outflow, juxta-anastomotic and inflow stenosis: ≥50% narrowing of the lumen anywhere along the access, recorded by location: outflow vein, juxta-anastomotic segment, or inflow artery. Tick each location where it is present.

Definitions follow the methods of Heindel et al., J Vasc Access 2025.

viewof baseline = baselineForm()

resultCard(baselineCurve)

responsive((w) => interventionsPlot(baselineCurve, w), invalidation)

The baseline model is a random survival forest. The curve and the 1-year value are the forest’s ensemble cumulative hazard.

How the inputs are defined

All characteristics are as of the day the AVF was created.

Tobacco use: Smoking status: never, current or former.
Diabetes: History of diabetes.
Heart failure: History of heart failure (congestive heart disease).
Anticoagulation: Taking an anticoagulant when the AVF was created: started on or before the day of surgery, not counting medications given only around the time of surgery. Included warfarin, apixaban, rivaroxaban, dabigatran, edoxaban, enoxaparin, fondaparinux, argatroban, bivalirudin and antithrombin.
Antiplatelet: Taking an antiplatelet agent when the AVF was created (same rule). Included aspirin, clopidogrel, prasugrel, ticagrelor, cilostazol and dipyridamole.
Statin: Taking a statin at baseline.
Prior transplant: History of kidney transplant.
On hemodialysis at AVF creation: Already receiving hemodialysis when the AVF was created, rather than not yet on dialysis.
Current or prior CVC: A central venous catheter now or in the past.
Vein and artery diameter: Measured in the operating room, before the arteriotomy and vein transection.
AVF configuration: Location of the radiocephalic anastomosis: wrist, proximal forearm, or anatomical snuffbox.
AVF side: Left or right arm.
Suture technique: Running or interrupted anastomotic suture.

Definitions follow the methods of Heindel et al., J Vasc Access 2025, and the PATENCY trial data used to develop the model.

import {
  coxExpected, intUltrasoundCovariates, intBaselineCovariates,
  INT_US_DEFAULTS, INT_BL_DEFAULTS, INT_TIMES, rsfPrepare, rsfPredict,
} from "./js/models.js"

coxEarly = FileAttachment("models/int_cox_early.json").json()
coxLate = FileAttachment("models/int_cox_late.json").json()
forest = FileAttachment("models/int_rsf_bl.json").json().then(rsfPrepare)

earlyCurve = coxExpected(coxEarly, intUltrasoundCovariates("early", early), INT_TIMES)
lateCurve = coxExpected(coxLate, intUltrasoundCovariates("late", late), INT_TIMES)
baselineCurve = rsfPredict(forest, intBaselineCovariates(baseline)).chf

grouped = (groups) => Inputs.form(
  Object.assign({}, ...groups.map(([, inputs]) => inputs)),
  {
    template: (inputs) => html`<div>${groups.map(([legend, group]) =>
      html`<fieldset><legend>${legend}</legend>${Object.keys(group).map((k) => inputs[k])}</fieldset>`)}</div>`,
  },
)

// Slider ranges, steps and defaults from the 2023 app (predict_int_shiny/app.R).
ultrasoundForm = (tab) => {
  const d = INT_US_DEFAULTS[tab];
  const tdos = tab === "early" ? [28, 56] : [70, 98];
  const maxDiam = tab === "early" ? 7 : 8;
  const diam = (label, value) => Inputs.range([2, maxDiam], { step: 0.5, value, label });
  return grouped([
    ["Timing and history", {
      tdos: Inputs.range(tdos, { step: 7, value: d.tdos, label: "Days since AVF creation at ultrasound" }),
      priorReints: Inputs.range([0, 4], { step: 1, value: d.priorReints, label: "Prior interventions" }),
      avfUsed: Inputs.toggle({ label: "AVF in use for dialysis", value: d.avfUsed }),
    }],
    ["Flow", {
      flow: Inputs.range([200, 2000], { step: 100, value: d.flow, label: "Flow volume (mL/min)" }),
    }],
    ["Access diameter (mm)", {
      distDiam: diam("Distal", d.distDiam),
      midDiam: diam("Mid", d.midDiam),
      proxDiam: diam("Proximal", d.proxDiam),
    }],
    ["Access stenosis", {
      outflowStenosis: Inputs.toggle({ label: "Outflow stenosis", value: d.outflowStenosis }),
      juxtaStenosis: Inputs.toggle({ label: "Juxta-anastomotic stenosis", value: d.juxtaStenosis }),
      inflowStenosis: Inputs.toggle({ label: "Inflow stenosis", value: d.inflowStenosis }),
    }],
  ]);
}

baselineForm = () => {
  const d = INT_BL_DEFAULTS;
  const radio = (label, options, value) => Inputs.radio(new Map(options), { label, value });
  const diam = (label, value) => Inputs.range([2, 5], { step: 0.5, value, label });
  return grouped([
    ["Demographics", {
      age: Inputs.range([30, 90], { step: 5, value: d.age, label: "Age (years)" }),
      sex: radio("Sex", [["Female", "female"], ["Male", "male"]], d.sex),
      tobacco: radio("Tobacco use", [["Never", "never"], ["Current", "current"], ["Former", "former"]], d.tobacco),
    }],
    ["Comorbidities", {
      diabetes: Inputs.toggle({ label: "Diabetes", value: d.diabetes }),
      heartFailure: Inputs.toggle({ label: "Heart failure", value: d.heartFailure }),
    }],
    ["Medications", {
      anticoagulant: Inputs.toggle({ label: "Anticoagulation", value: d.anticoagulant }),
      antiplatelet: Inputs.toggle({ label: "Antiplatelet", value: d.antiplatelet }),
      statin: Inputs.toggle({ label: "Statin", value: d.statin }),
    }],
    ["Kidney disease history", {
      priorTransplant: Inputs.toggle({ label: "Prior transplant", value: d.priorTransplant }),
      onHemodialysis: Inputs.toggle({ label: "On hemodialysis at AVF creation", value: d.onHemodialysis }),
      cvc: Inputs.toggle({ label: "Current or prior CVC", value: d.cvc }),
    }],
    ["Anatomy", {
      veinDiam: diam("Vein diameter (mm)", d.veinDiam),
      artDiam: diam("Artery diameter (mm)", d.artDiam),
    }],
    ["Operative plan", {
      location: radio("AVF configuration", [["Wrist", "wrist"], ["Forearm", "forearm"], ["Snuffbox", "snuffbox"]], d.location),
      side: radio("AVF side", [["Left", "left"], ["Right", "right"]], d.side),
      technique: radio("Suture technique", [["Running", "running"], ["Interrupted", "interrupted"]], d.technique),
    }],
  ]);
}

resultCard = (curve) => html`<div class="result-card">
  <div class="result-value">${curve.at(-1).toFixed(2)}</div>
  <div class="result-label">expected interventions in the first year</div>
  <div class="result-note">Cumulative by day 360 after AVF creation.</div>
</div>`

// KDOQI 2019 "1-2-3" intervention goal (Lok et al., AJKD 2020, Goals and Targets, S133):
// <=2 interventions to facilitate AV access use; <=3 per year to maintain use.
// Short labels for narrow (phone) plots so they don't run off the edge.
KDOQI = [
  { y: 2, label: "KDOQI goal: ≤2 interventions to facilitate use", short: "KDOQI: ≤2 to facilitate use" },
  { y: 3, label: "KDOQI goal: ≤3 per year to maintain use", short: "KDOQI: ≤3/yr to maintain use" },
]

// Re-render at the container's width (so text stays legible on phones instead of a
// scaled-down SVG). Hidden tabs have width 0 and render when first shown.
responsive = (render, invalidation) => {
  const div = html`<div style="width: 100%"></div>`;
  let current = 0;
  const observer = new ResizeObserver(([entry]) => {
    const w = Math.floor(entry.contentRect.width);
    if (w > 0 && w !== current) {
      current = w;
      div.replaceChildren(render(w));
    }
  });
  observer.observe(div);
  invalidation.then(() => observer.disconnect());
  return div;
}

// Step curve of expected interventions, with crosshairs where it first reaches each
// whole number and the KDOQI thresholds as reference lines.
interventionsPlot = (values, width) => {
  const data = INT_TIMES.map((t, i) => ({ t, y: values[i] }));
  const max = d3.max(values);
  const reached = d3.range(1, Math.floor(max) + 1).map((k) => ({ k, t: INT_TIMES[values.findIndex((v) => v >= k)] }));
  return Plot.plot({
    width,
    height: 340,
    marginLeft: 44,
    marginBottom: 46,
    style: { fontSize: "13px", background: "transparent" },
    x: {
      domain: [0, 360], ticks: [0, 90, 180, 270, 360],
      label: "Days since AVF creation", labelAnchor: "center", labelArrow: "none", labelOffset: 40,
    },
    y: { domain: [0, Math.max(3.5, Math.ceil(max) + 0.5)], grid: true, label: "Expected interventions" },
    marks: [
      Plot.ruleY(KDOQI, { y: "y", stroke: "var(--pa-kdoqi)", strokeDasharray: "5,4" }),
      Plot.text(KDOQI, { x: 4, y: "y", text: width < 520 ? "short" : "label", textAnchor: "start", dy: -8, fill: "var(--pa-kdoqi)" }),
      Plot.ruleY([0]),
      Plot.ruleX(reached, { x: "t", y1: 0, y2: "k", strokeOpacity: 0.3 }),
      Plot.ruleY(reached, { y: "k", x1: 0, x2: "t", strokeOpacity: 0.3 }),
      Plot.line(data, { x: "t", y: "y", curve: "step-after", stroke: "var(--bs-primary)", strokeWidth: 3 }),
      Plot.dot(reached, { x: "t", y: "k", fill: "var(--bs-primary)", r: 4 }),
      Plot.tip(data, Plot.pointerX({
        x: "t", y: "y",
        title: (d) => `Day ${d.t}: ${d.y.toFixed(2)} expected interventions`,
      })),
    ],
  });
}