表型 · 本地处理

Audit Manifest Viewer

把 audit_write() 的 JSON 读成一页溯源记录：取了哪些字段、每一步剩多少人、拟了什么模型；比较两份 manifest 的差别，并导出 CONSORT 流程图的行与投稿用的摘要。

六层记录逐条摆开

manifest

选择文件

{
  "schema_version": 1,
  "meta": {
    "name": "smoking_lung_cancer",
    "start_time": "2026-08-11T18:57:24+0800",
    "ukbflow_version": "0.4.0",
    "r_version": "4.4.1",
    "platform": "x86_64-w64-mingw32",
    "dx_user": null,
    "dx_project": null,
    "session_info": [
      "R version 4.4.1 (2024-06-14)",
      "Platform: x86_64-w64-mingw32/x64",
      "Running under: Windows 11 x64 (build 26200)",
      "attached base packages: stats graphics grDevices utils datasets methods base",
      "other attached packages: ukbflow_0.4.0 data.table_1.15.4 survival_3.7-0"
    ]
  },
  "fields": [
    {
      "label": "analysis_fields",
      "field_id": [31, 53, 21022, 21001, 20116, 1558, 22189, 54, 22009, 20001, 20006, 40006, 40011, 40012, 40005, 40000],
      "dataset": null,
      "n_fields": 16,
      "recorded_at": "2026-08-11T18:57:31+0800"
    }
  ],
  "recipes": [],
  "snapshots": [
    {
      "label": "raw",
      "nrow": 500000,
      "ncol": 24,
      "n_na_cols": 11,
      "columns": ["eid", "sex", "p34", "date_of_attending_assessment_centre_i0", "age_at_recruitment", "body_mass_index_bmi_i0", "smoking_status_i0", "alcohol_intake_frequency_i0", "townsend_deprivation_index_at_recruitment", "uk_biobank_assessment_centre_i0", "genetic_principal_components", "bmi_cat", "tdi_cat", "smoking_ever", "lung_selfreport", "lung_selfreport_date", "lung_icd10", "lung_icd10_date", "lung_status", "lung_date", "lung_timing", "lung_followup_end", "lung_followup_years", "date_of_death"],
      "object_size_mb": 214.6,
      "recorded_at": "2026-08-11T19:02:47+0800"
    },
    {
      "label": "after excluding prevalent cases",
      "nrow": 494955,
      "ncol": 24,
      "n_na_cols": 11,
      "columns": ["eid", "sex", "p34", "date_of_attending_assessment_centre_i0", "age_at_recruitment", "body_mass_index_bmi_i0", "smoking_status_i0", "alcohol_intake_frequency_i0", "townsend_deprivation_index_at_recruitment", "uk_biobank_assessment_centre_i0", "genetic_principal_components", "bmi_cat", "tdi_cat", "smoking_ever", "lung_selfreport", "lung_selfreport_date", "lung_icd10", "lung_icd10_date", "lung_status", "lung_date", "lung_timing", "lung_followup_end", "lung_followup_years", "date_of_death"],
      "object_size_mb": 212.4,
      "recorded_at": "2026-08-11T19:03:12+0800"
    },
    {
      "label": "after excluding missing covariates",
      "nrow": 465927,
      "ncol": 24,
      "n_na_cols": 3,
      "columns": ["eid", "sex", "p34", "date_of_attending_assessment_centre_i0", "age_at_recruitment", "body_mass_index_bmi_i0", "smoking_status_i0", "alcohol_intake_frequency_i0", "townsend_deprivation_index_at_recruitment", "uk_biobank_assessment_centre_i0", "genetic_principal_components", "bmi_cat", "tdi_cat", "smoking_ever", "lung_selfreport", "lung_selfreport_date", "lung_icd10", "lung_icd10_date", "lung_status", "lung_date", "lung_timing", "lung_followup_end", "lung_followup_years", "date_of_death"],
      "object_size_mb": 199.9,
      "recorded_at": "2026-08-11T19:03:40+0800"
    }
  ],
  "phenotypes": [
    {
      "label": "lung",
      "name": "lung",
      "n": 500000,
      "recorded_at": "2026-08-11T19:01:55+0800",
      "selfreport": { "present": true, "status_col": "lung_selfreport", "date_col": "lung_selfreport_date", "n_cases": 1712, "n_dated": 1655 },
      "icd10": { "present": true, "status_col": "lung_icd10", "date_col": "lung_icd10_date", "n_cases": 8431, "n_dated": 8431 },
      "icd9": { "present": false, "status_col": "NA", "date_col": "NA", "n_cases": "NA", "n_dated": "NA" },
      "algorithm": { "present": false, "status_col": "NA", "date_col": "NA", "n_cases": "NA", "n_dated": "NA" },
      "sources": {
        "hes": { "present": false, "status_col": "NA", "date_col": "NA", "n_cases": "NA", "n_dated": "NA" },
        "hes_icd9": { "present": false, "status_col": "NA", "date_col": "NA", "n_cases": "NA", "n_dated": "NA" },
        "death": { "present": false, "status_col": "NA", "date_col": "NA", "n_cases": "NA", "n_dated": "NA" },
        "first_occurrence": { "present": false, "status_col": "NA", "date_col": "NA", "n_cases": "NA", "n_dated": "NA" },
        "cancer_registry": { "present": false, "status_col": "NA", "date_col": "NA", "n_cases": "NA", "n_dated": "NA" },
        "cancer_registry_icd9": { "present": false, "status_col": "NA", "date_col": "NA", "n_cases": "NA", "n_dated": "NA" }
      },
      "combined": { "present": true, "status_col": "lung_status", "date_col": "lung_date", "n_cases": 9085, "n_dated": 9085 },
      "timing": { "present": true, "timing_col": "lung_timing", "no_disease": 490915, "prevalent": 5045, "incident": 4040, "missing": 0 },
      "followup": { "present": true, "end_col": "lung_followup_end", "years_col": "lung_followup_years", "n_end": 500000, "n_non_missing": 500000, "mean": 13.12, "median": 13.64, "min": 0.01, "max": 16.82 }
    }
  ],
  "models": [
    {
      "label": "smoking_lung_cox",
      "method": "coxph",
      "n_rows": 3,
      "exposures": ["smoking_ever"],
      "exposure_levels": { "smoking_ever": ["Never", "Ever"] },
      "exposure_scores": {},
      "models": ["Unadjusted", "Age and sex adjusted", "Fully adjusted"],
      "outcome_col": "lung_status",
      "time_col": "lung_followup_years",
      "covariates": ["age_at_recruitment", "sex", "bmi_cat", "tdi_cat", "alcohol_intake_frequency_i0", "uk_biobank_assessment_centre_i0", "pc1", "pc2", "pc3", "pc4", "pc5", "pc6", "pc7", "pc8", "pc9", "pc10"],
      "strata": [],
      "cluster_col": "NA",
      "test": "wald",
      "ci_method": "profile",
      "conf_level": 0.95,
      "knots": [],
      "ref": "NA",
      "compete_col": "NA",
      "event_val": "NA",
      "compete_val": "NA",
      "results": [
        { "term": "smoking_everEver", "model": "Unadjusted", "HR": 1.04, "CI_lower": 0.98, "CI_upper": 1.11, "p": 0.203, "n": 465927, "n_events": 3762 },
        { "term": "smoking_everEver", "model": "Age and sex adjusted", "HR": 1.04, "CI_lower": 0.98, "CI_upper": 1.11, "p": 0.201, "n": 465927, "n_events": 3762 },
        { "term": "smoking_everEver", "model": "Fully adjusted", "HR": 1.04, "CI_lower": 0.98, "CI_upper": 1.11, "p": 0.201, "n": 465927, "n_events": 3762 }
      ],
      "recorded_at": "2026-08-11T19:08:02+0800"
    }
  ],
  "jobs": []
}

smoking_lung_cancer

schema v1

开始: 2026-08-11T18:57:24+0800
ukbflow: 0.4.0
R: 4.4.1
平台: x86_64-w64-mingw32
dx 用户: —
dx 项目: —

1字段0配方3快照1表型1模型0作业

analysis_fields16 个字段

2026-08-11 18:57

3153210222100120116155822189542200920001200064000640011400124000540000

这一层没有记录。

标签	行	列	含 NA 的列	大小	记录时间
raw	500,000	24	11	214.6 MB	2026-08-11 19:02
after excluding prevalent cases	494,955	24	11	212.4 MB	2026-08-11 19:03
after excluding missing covariates	465,927	24	3	199.9 MB	2026-08-11 19:03

lung队列 500,000 人

2026-08-11 19:01

各来源命中的病例数

自报1,712

ICD-108,431

合并9,085

相对基线的时序

无该病none490,915

基线前已患病prevalent5,045

随访期新发incident4,040

随访时间

中位数13.64 年

最短0.01 年

最长16.82 年

smoking_lung_coxcoxph

2026-08-11 19:08

结局: lung_status
随访时间: lung_followup_years
暴露: smoking_ever
协变量: age_at_recruitmentsexbmi_cattdi_catalcohol_intake_frequency_i0uk_biobank_assessment_centre_i0pc1pc2pc3pc4pc5pc6pc7pc8pc9pc10
调整层次: UnadjustedAge and sex adjustedFully adjusted

检验wald

区间法profile

置信水平95%

结果行3

这一层没有记录。

说明

manifest 是什么: audit_start() 起一份记录，audit_fields()、audit_snapshot()、audit_model() 等在分析途中往里追加，audit_write() 落成 JSON。它记的是一次分析的来历：取了哪些字段、用哪一版配方、每一步剩多少人、拟了什么模型、跑过哪些作业。
为什么可以带出 RAP: manifest 全部是汇总级的 —— 行数、列名、字段号、效应量，没有任何个体行。所以它能带出 RAP、发给合作者、附进投稿材料，而个体数据留在原地。这个页面同样在本地解析，文件不上传。
三种视图: 「清单」对应 summary() 在控制台给的东西；「流转」对应 audit_diff(aud, "snapshots") 这类层内比较；「比较」对应 audit_diff(aud, audit2 = )。
流程图那五列: audit_flowchart() 的输出格式：id / parent / n / label / type。快照层本来就是一条流失链，差额作为 exclusion 兄弟行插在每一步之前 —— 兄弟而非父子，排除框才落在主干旁边。复制过去粘进 Flowchart Builder 就是一张 CONSORT 图。
缺失值的两副面孔: R 的 NA 写进 JSON 后，顶层字段是 null，而嵌套在记录里的（如表型的 n_cases）会成为字符串 "NA"。两种这里都当作缺失，所以不会看到一列写着 "NA" 的字符串。
校验的分寸: 错误是「这不是一份 manifest」（骨架缺失、记录层不是数组）；警告是「读得下去但不太对」（label 重复、schema 版本更新）。R 那边 audit_write() 会在写入前一律拦下，而这个页面的用处恰恰是把一份已经写坏的文件看明白，所以只提示、不拒绝。