{
  "docId": "019dd923-5ca1-7489-b638-e91d3b8478cc",
  "docSlug": "9fbedbfea0231cbe",
  "documentTitle": "AI Healthcare Errors",
  "authorId": "McKinsey",
  "authorName": "McKinsey",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "strategy_consulting",
  "sourceTypeLabel": "Strategy consulting",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.333,
  "pageNumber": 10,
  "pageCount": 26,
  "prevPage": 9,
  "nextPage": 11,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "overcrowded",
  "nDataPoints": 12,
  "notes": "The slide uses a hub-and-spoke diagram to illustrate data integration and bar charts to compare model performance across different data inputs.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "infographic",
    "bar_chart_vertical",
    "footnote"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5ca1-7489-b638-e91d3b8478cc/10",
  "deckHref": "/decks/019dd923-5ca1-7489-b638-e91d3b8478cc",
  "deckJsonHref": "/decks/019dd923-5ca1-7489-b638-e91d3b8478cc.json",
  "deckAnchorHref": "/decks/019dd923-5ca1-7489-b638-e91d3b8478cc#slide-10",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Clinical coding data has higher predictive performance than clinical measurements – though both used in combination have higher predictive performance for certain ADEs.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd951-d3bc-73c4-a2ba-13e796bb7abd",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.45,
        "x": 0.52,
        "y": 0.0065
      },
      "kind": "chart",
      "text": "AUC (area under curve)",
      "attrs": null,
      "subkind": "bar-vertical",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "1703e159-15e7-43fd-95ca-35c014380bb4",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.45,
        "x": 0.03,
        "y": 0.0065
      },
      "kind": "chart",
      "text": "Accuracy score",
      "attrs": null,
      "subkind": "bar-vertical",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "41797252-2691-4924-be24-e7a70ab4fa75",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.45,
        "x": 0.52,
        "y": 0.0015
      },
      "kind": "diagram",
      "text": "Extracting data for machine learning from EHRs",
      "attrs": null,
      "subkind": "hub-spoke",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "04cdfd1a-72d9-43b2-a307-a4fa6a3eda06",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.45,
        "x": 0.03,
        "y": 0.0015
      },
      "kind": "list",
      "text": "Machine learning (random forest model) can be used to accurately predict ADEs (adverse drug events) using data from EHRs (electronic health records). Clinical coding data has higher predictive performance than clinical measurements. Feature selection improves performance. ADEs are ~5% of hospital admissions. Voluntary reporting misses ~94% of ADEs.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "00c688bb-0e27-41e1-a645-68c827903e6c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "AUC: 0.76",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd951-d3bc-73c4-a2ba-16b151b72343",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.7,
        "x": 0.03,
        "y": 0.95
      },
      "kind": "source-note",
      "text": "SOURCE: Zhoa et al, Predictive modeling of structured electronic health records for adverse drug event detection, 2015, BMC Medical Informatics and Decision Making, 15 (Supplement 4):S1",
      "attrs": null,
      "subkind": null,
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "d22c4b98-ded2-4989-9e2f-dd58e647fae9",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.9,
        "x": 0.03,
        "y": 0.03
      },
      "kind": "title",
      "text": "Proof of concept established for machine learning based predictive modelling to detect adverse drug events using EHR data",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "669f1777-80de-4f86-88a0-fb4d61f647a6",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [
    {
      "metricRaw": "AUC",
      "numberRaw": "0.76",
      "numberKind": "plain",
      "actionTitle": "Proof of concept established for machine learning based predictive modelling to detect adverse drug events using EHR data",
      "calloutText": "Clinical coding data has higher predictive performance than clinical measurements – though both used in combination have higher predictive performance for certain ADEs.",
      "numberScale": null,
      "numberValue": 0.76,
      "metricFamily": "other",
      "numberCurrency": null
    }
  ],
  "tools": [],
  "frameworks": [
    {
      "name": "hub-spoke",
      "slug": null,
      "matchId": "ab2100dd-ee98-4828-afe1-f2a6a98c3519",
      "evidence": "Diagram showing data sources feeding into a central patient record",
      "confidence": 0.9
    }
  ],
  "arcBeats": [
    {
      "to": 26,
      "from": 10,
      "beatId": "9b6e4ffe-95ca-40d4-92d5-70f2cd8f60f8",
      "arcName": "Problem-Agitate-Solution",
      "arcSlug": "problem-agitate-solution",
      "beatName": "Solution (Provide relief)",
      "beatSlug": "problem-agitate-solution-solution-provide-relief",
      "evidence": "The deck offers AI as a solution to prevent healthcare errors and improve patient outcomes.",
      "position": 2,
      "confidence": 0.8,
      "parentBeatName": "Resolution",
      "parentBeatSlug": "resolution"
    }
  ],
  "loops": [
    {
      "to": 14,
      "from": 4,
      "name": "Golden Circle",
      "slug": "11-golden-circle",
      "bestFor": "Visionary leadership, brand positioning, mission statements",
      "matchId": "1b36e03c-871f-4ba9-977e-2461066df8bf",
      "evidence": "The deck presents various case studies and examples of AI applications in healthcare, using the golden circle framework.",
      "position": 1,
      "objective": "Explain why AI is different and how it can be applied in healthcare",
      "structure": "The Why (Belief) -> The How (Process) -> The What (Result)",
      "confidence": 0.6,
      "description": "Invert the typical pitch by starting with why you exist, rather than what you do"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}