Spaces:
Sleeping
Sleeping
File size: 45,980 Bytes
7a12bf7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 | #!/usr/bin/env python3
"""Rule-based 30-day readmission risk classification engine.
Reference implementation of the algorithm described in ALGORITHM_DESIGN.md.
Input: TOON lines (CLUSTER|Keyword|Value|Timestamp)
Output: Risk classification + days-to-readmission prediction
Usage:
# From TOON string
engine = ReadmissionRiskEngine()
result = engine.score_from_toon(toon_text)
print(result)
# From TOON file
result = engine.score_from_file("path/to/extraction.txt")
# From JSONL training data
results = engine.score_from_jsonl("dspy_fine_tuning/data/trainset_full.jsonl")
"""
from __future__ import annotations
import json
import math
import re
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple, Union
# ---------------------------------------------------------------------------
# Data classes
# ---------------------------------------------------------------------------
@dataclass
class ParsedFact:
cluster: str
keyword: str
value: Union[float, str]
timestamp: str
is_numeric: bool
plausibility_ok: bool = True
@dataclass
class ClusterScore:
cluster: str
score: int
max_score: int
contributing_factors: List[str] = field(default_factory=list)
@dataclass
class InteractionResult:
pattern_id: str
pattern_name: str
bonus: int
description: str
@dataclass
class SurvivalCurve:
"""P(readmit by day t) for several horizons."""
horizons: Dict[int, float] # {7: 0.05, 14: 0.12, 21: 0.18, 30: 0.23}
@dataclass
class RiskResult:
# Scores
composite_score: int
cluster_scores: Dict[str, ClusterScore]
interaction_bonus: int
interactions_triggered: List[InteractionResult]
# Risk classification
probability: float
risk_category: str # Low / Medium / High / Critical
risk_color: str
# Days prediction
estimated_days: float
days_bucket: str # "0-7 days" / "8-14 days" / "15-30 days"
survival_curve: SurvivalCurve
# Explainability
risk_factors: List[str]
protective_factors: List[str]
missing_clusters: List[str]
data_completeness: float
confidence: str # high / medium / low
# Raw data
n_facts_parsed: int
n_facts_dropped: int
# ---------------------------------------------------------------------------
# Constants
# ---------------------------------------------------------------------------
VALID_CLUSTERS = {
"DEMOGRAPHICS", "VITALS", "LABS", "PROBLEMS", "SYMPTOMS",
"MEDICATIONS", "PROCEDURES", "UTILIZATION", "DISPOSITION",
}
NUMERIC_CLUSTERS = {"VITALS", "LABS", "UTILIZATION"}
OBJECTIVE_CLUSTERS = {"DEMOGRAPHICS", "VITALS", "LABS", "UTILIZATION", "DISPOSITION"}
# ---------------------------------------------------------------------------
# Engine
# ---------------------------------------------------------------------------
class ReadmissionRiskEngine:
"""Main entry point for readmission risk scoring."""
def __init__(self, config_dir: Optional[Path] = None):
if config_dir is None:
config_dir = Path(__file__).parent / "config"
self._config_dir = config_dir
self._scoring_rules = self._load_json("scoring_rules.json")
self._problem_groups = self._load_json("snomed_problem_groups.json")["groups"]
self._symptom_groups = self._load_json("symptom_urgency_groups.json")["groups"]
# Build lookup indexes
self._problem_synonym_index = self._build_synonym_index(self._problem_groups)
self._symptom_synonym_index = self._build_synonym_index(self._symptom_groups)
# Calibration parameters
cal = self._scoring_rules["_meta"]["calibration"]
self._alpha = cal["alpha"]
self._beta = cal["beta"]
# Days prediction parameters
days_cfg = self._scoring_rules["DAYS_PREDICTION"]["models"]
reg = days_cfg["regression"]["parameters"]
self._d_max = reg["D_max"]
self._gamma = reg["gamma"]
surv = days_cfg["survival"]["parameters"]
self._k_base = surv["k_base"]
# -- Loading helpers ----------------------------------------------------
def _load_json(self, filename: str) -> Dict[str, Any]:
p = self._config_dir / filename
return json.loads(p.read_text(encoding="utf-8"))
@staticmethod
def _build_synonym_index(groups: List[Dict]) -> Dict[str, str]:
"""Map lowercase synonym → group id."""
idx: Dict[str, str] = {}
for g in groups:
gid = g["id"]
for syn in g.get("synonyms", []):
key = syn.strip().lower()
if key not in idx:
idx[key] = gid
return idx
def _match_to_group(
self,
keyword: str,
synonym_index: Dict[str, str],
groups: List[Dict],
) -> Optional[Dict]:
"""Smart matching: exact > word-boundary substring > raw substring.
Avoids false matches like 'tia' in 'essential' by preferring
word-boundary matches and longer synonyms.
"""
kw_lower = keyword.strip().lower()
# 1) Exact match (full keyword == synonym)
gid = synonym_index.get(kw_lower)
if gid:
return self._group_by_id(groups, gid)
# Tokenize keyword into words for word-boundary matching
kw_words = set(re.split(r"[\s,;/\-()]+", kw_lower))
# 2) Word-boundary match: synonym is a whole word within the keyword
# OR keyword starts/ends with the synonym as a distinct token
best_wb_match: Optional[str] = None
best_wb_len = 0
# 3) Raw substring match (fallback, requires min 4 chars to avoid noise)
best_sub_match: Optional[str] = None
best_sub_len = 0
for syn, gid in synonym_index.items():
if syn not in kw_lower:
continue
# Check if it's a word-boundary match
is_word_match = (
syn in kw_words # exact word token
or kw_lower.startswith(syn + " ")
or kw_lower.endswith(" " + syn)
or (" " + syn + " ") in kw_lower
)
if is_word_match and len(syn) > best_wb_len:
best_wb_match = gid
best_wb_len = len(syn)
elif not is_word_match and len(syn) >= 4 and len(syn) > best_sub_len:
# Only use raw substring for synonyms >= 4 chars
best_sub_match = gid
best_sub_len = len(syn)
# Prefer word-boundary matches over raw substring
chosen = best_wb_match or best_sub_match
if chosen:
return self._group_by_id(groups, chosen)
return None
# -- Layer 1: Parser & Normalizer ----------------------------------------
@staticmethod
def _try_parse_float(value: str) -> Optional[float]:
"""Best-effort numeric parse.
Stage2 should emit numeric-only values for numeric fields, but in practice
we sometimes see light decoration like '3 days'. For scoring purposes we
accept the first numeric token, but we avoid parsing ratios like '120/80'.
"""
s = (value or "").strip()
if not s:
return None
# Avoid BP-style ratios and similar formats.
if "/" in s:
return None
# Fast path: pure float
try:
return float(s)
except Exception:
pass
# Fallback: extract first numeric token
m = re.search(r"[-+]?\d+(?:\.\d+)?", s)
if not m:
return None
try:
return float(m.group(0))
except Exception:
return None
@staticmethod
def _split_semantic_items(value: str, *, limit: int = 20) -> List[str]:
"""Split a semicolon/comma/newline separated list into normalized items."""
raw = (value or "").strip()
if not raw:
return []
parts: List[str] = []
for seg in re.split(r"[;\n]+", raw):
seg = seg.strip()
if not seg:
continue
for item in seg.split(","):
it = " ".join(item.strip().split())
if not it:
continue
parts.append(it.strip(" -"))
if len(parts) >= limit:
break
if len(parts) >= limit:
break
# Dedup while preserving order.
out: List[str] = []
seen: set[str] = set()
for it in parts:
k = it.casefold()
if k in seen:
continue
seen.add(k)
out.append(it)
return out
@staticmethod
def _strip_prefix(keyword: str, prefixes: List[str]) -> str:
k = (keyword or "").strip()
k_cf = k.casefold()
for p in prefixes:
p_cf = p.casefold()
if k_cf.startswith(p_cf):
k = k[len(p) :].strip()
k_cf = k.casefold()
return k
@staticmethod
def _normalize_discharge_disposition(value: str) -> str:
"""Normalize common discharge disposition variants to the scoring allowlist."""
v = (value or "").strip()
v_cf = v.casefold()
if not v:
return v
# Canonical allowlist (scoring_rules.json): Home, Home with Services, Rehab, SNF, LTAC, Hospice, AMA
if v_cf in {"home with service", "home w service", "home with svc", "home w/ service"}:
return "Home with Services"
if v_cf in {"home with services", "home w services", "home w/ services", "home health", "home health care"}:
return "Home with Services"
if v_cf in {"hospice residence", "hospice care"}:
return "Hospice"
return v
@staticmethod
def _normalize_mental_status(value: str) -> str:
v = (value or "").strip()
v_cf = v.casefold()
if not v:
return v
if "alert" in v_cf and "orient" in v_cf:
return "alert"
if v_cf in {"a&o", "ao", "a/ox3", "a/ox4"}:
return "alert"
return v
def parse_toon(self, toon_text: str) -> Tuple[Dict[str, List[ParsedFact]], int, int]:
"""Parse TOON text into structured facts.
Returns (facts_by_cluster, n_parsed, n_dropped).
"""
facts: Dict[str, List[ParsedFact]] = {}
n_parsed = 0
n_dropped = 0
seen_objective: set = set()
for raw_line in toon_text.strip().splitlines():
line = raw_line.strip()
if not line or line.startswith("#"):
continue
parts = line.split("|")
if len(parts) != 4:
n_dropped += 1
continue
cluster, keyword, value, timestamp = (p.strip() for p in parts)
if cluster not in VALID_CLUSTERS:
n_dropped += 1
continue
# Strip common semantic prefixes embedded in the keyword.
if cluster == "PROBLEMS":
keyword = self._strip_prefix(keyword, ["PMH:", "PMH/Comorbidities:", "Discharge Dx:", "Working Dx:", "Complication:", "Complications:"])
elif cluster == "SYMPTOMS":
keyword = self._strip_prefix(keyword, ["ADM:", "DC:"])
# Expand common Stage2 aggregate semantic lines into per-item facts.
# This makes the scorer robust to model drift like:
# PROBLEMS|Discharge Dx|CHF; COPD|Discharge
# instead of emitting one line per diagnosis.
if cluster == "PROBLEMS":
kw_cf = keyword.strip().casefold()
acute_keys = {"discharge dx", "working dx", "complication", "complications"}
chronic_keys = {"pmh/comorbidities", "pmh", "comorbidities", "past medical history"}
items = self._split_semantic_items(value)
if kw_cf in acute_keys and items:
for it in items:
fact = ParsedFact(
cluster="PROBLEMS",
keyword=it,
value="acute",
timestamp="Discharge",
is_numeric=False,
plausibility_ok=True,
)
facts.setdefault("PROBLEMS", []).append(fact)
n_parsed += 1
continue
if kw_cf in chronic_keys and items:
for it in items:
fact = ParsedFact(
cluster="PROBLEMS",
keyword=it,
value="chronic",
timestamp="Past",
is_numeric=False,
plausibility_ok=True,
)
facts.setdefault("PROBLEMS", []).append(fact)
n_parsed += 1
continue
# Numeric parsing:
# - Strictly numeric clusters MUST parse (else drop).
# - Non-numeric clusters may still have numeric keywords (e.g. MEDICATIONS Medication Count,
# PROCEDURES Mechanical Ventilation days). Those should parse so scoring rules apply.
is_numeric = False
parsed_value: Union[float, str] = value
kw_rules = self._scoring_rules.get(cluster, {}).get("keywords", {}).get(keyword, {})
kw_type = kw_rules.get("type") if isinstance(kw_rules, dict) else None
if cluster in NUMERIC_CLUSTERS:
v = self._try_parse_float(value)
if v is None:
n_dropped += 1
continue
parsed_value = v
is_numeric = True
elif kw_type == "range":
v = self._try_parse_float(value)
if v is None:
n_dropped += 1
continue
parsed_value = v
is_numeric = True
elif kw_type == "mixed":
# Mixed: numeric is optional; keep as string if parsing fails.
v = self._try_parse_float(value)
if v is not None:
parsed_value = v
is_numeric = True
# Plausibility check
plausibility_ok = True
if is_numeric:
plausibility_ok = self._check_plausibility(cluster, keyword, parsed_value)
# Dedup for objective clusters
if cluster in OBJECTIVE_CLUSTERS:
key = (cluster, keyword)
if key in seen_objective:
# Keep the one with better timestamp
n_dropped += 1
continue
seen_objective.add(key)
fact = ParsedFact(
cluster=cluster,
keyword=keyword,
value=parsed_value,
timestamp=timestamp,
is_numeric=is_numeric,
plausibility_ok=plausibility_ok,
)
facts.setdefault(cluster, []).append(fact)
n_parsed += 1
return facts, n_parsed, n_dropped
def _check_plausibility(self, cluster: str, keyword: str, value: float) -> bool:
cluster_rules = self._scoring_rules.get(cluster, {}).get("keywords", {})
kw_rules = cluster_rules.get(keyword, {})
plaus = kw_rules.get("plausibility")
if plaus:
return plaus["min"] <= value <= plaus["max"]
return True
# -- Layer 2: Concept Mapper --------------------------------------------
def map_problem_to_group(self, keyword: str) -> Optional[Dict]:
"""Map a PROBLEMS keyword to a SNOMED concept group."""
return self._match_to_group(keyword, self._problem_synonym_index, self._problem_groups)
def map_symptom_to_group(self, keyword: str) -> Optional[Dict]:
"""Map a SYMPTOMS keyword to an urgency group."""
return self._match_to_group(keyword, self._symptom_synonym_index, self._symptom_groups)
@staticmethod
def _group_by_id(groups: List[Dict], gid: str) -> Optional[Dict]:
for g in groups:
if g["id"] == gid:
return g
return None
# -- Layer 3: Cluster Scorers -------------------------------------------
def _score_range_keyword(self, rules: Dict, value: float) -> Tuple[int, str]:
"""Score a numeric value using range rules. Returns (score, label)."""
for r in rules.get("ranges", []):
if r["min"] <= value <= r["max"]:
return r["score"], r.get("label", "")
return 0, ""
def score_demographics(self, facts: List[ParsedFact]) -> ClusterScore:
rules = self._scoring_rules["DEMOGRAPHICS"]["keywords"]
score = 0
factors: List[str] = []
age_found = False
for f in facts:
if f.keyword == "Age" and f.is_numeric:
age_found = True
pts, label = self._score_range_keyword(rules["Age"], f.value)
score += pts
if pts > 0:
factors.append(f"Age {int(f.value)} ({label}, +{pts})")
elif f.keyword == "Sex":
val = str(f.value).lower()
pts = rules["Sex"]["values"].get(val, 0)
score += pts
if pts > 0:
factors.append(f"Sex={val} (+{pts})")
if not age_found:
default = rules["Age"].get("missing_score", 2)
score += default
factors.append(f"Age missing (default +{default})")
return ClusterScore("DEMOGRAPHICS", score, 10, factors)
def score_vitals(self, facts: List[ParsedFact]) -> ClusterScore:
rules = self._scoring_rules["VITALS"]["keywords"]
score = 0
factors: List[str] = []
for f in facts:
if not f.is_numeric or not f.plausibility_ok:
continue
kw_rules = rules.get(f.keyword)
if not kw_rules or kw_rules.get("type") == "no_direct_score":
continue
pts, label = self._score_range_keyword(kw_rules, f.value)
score += pts
if pts > 0:
factors.append(f"{f.keyword}={f.value} ({label}, +{pts})")
return ClusterScore("VITALS", score, 25, factors)
def score_labs(self, facts: List[ParsedFact]) -> ClusterScore:
rules = self._scoring_rules["LABS"]["keywords"]
score = 0
factors: List[str] = []
for f in facts:
if not f.is_numeric or not f.plausibility_ok:
continue
kw_rules = rules.get(f.keyword)
if not kw_rules:
continue
pts, label = self._score_range_keyword(kw_rules, f.value)
score += pts
if pts > 0:
factors.append(f"{f.keyword}={f.value} ({label}, +{pts})")
return ClusterScore("LABS", score, 30, factors)
def score_problems(self, facts: List[ParsedFact]) -> ClusterScore:
score = 0
factors: List[str] = []
active_groups: Dict[str, int] = {} # group_id -> max weight
include_values = {"chronic", "acute", "exist"}
for f in facts:
val = str(f.value).lower().strip()
if val not in include_values:
continue
group = self.map_problem_to_group(f.keyword)
if group:
gid = group["id"]
w = group["risk_weight"]
if gid not in active_groups or w > active_groups[gid]:
active_groups[gid] = w
factors.append(f"{f.keyword} → {group['name']} (weight {w})")
base_score = sum(active_groups.values())
# Multimorbidity bonus
n_groups = len(active_groups)
mm_bonus = 0
if n_groups > 3:
mm_bonus = min(n_groups - 3, 5)
factors.append(f"Multimorbidity: {n_groups} groups (+{mm_bonus})")
score = min(base_score + mm_bonus, 40)
return ClusterScore("PROBLEMS", score, 40, factors)
def score_symptoms(self, facts: List[ParsedFact]) -> ClusterScore:
sev_mult = {"severe": 1.5, "yes": 1.0, "no": 0.0}
score = 0.0
factors: List[str] = []
active_groups: Dict[str, float] = {}
active_count = 0
for f in facts:
val = str(f.value).lower().strip()
mult = sev_mult.get(val, 0.0)
if mult == 0.0:
continue
active_count += 1
group = self.map_symptom_to_group(f.keyword)
if group:
gid = group["id"]
w = group["risk_weight"] * mult
if gid not in active_groups or w > active_groups[gid]:
active_groups[gid] = w
factors.append(f"{f.keyword}={val} → {group['name']} (+{w:.1f})")
base_score = sum(active_groups.values())
# Active symptom count bonus
bonus = 0
if active_count > 3:
bonus = 2
factors.append(f"Active symptoms: {active_count} (>3, +2)")
score = min(int(round(base_score + bonus)), 15)
return ClusterScore("SYMPTOMS", score, 15, factors)
def score_medications(self, facts: List[ParsedFact]) -> ClusterScore:
rules = self._scoring_rules["MEDICATIONS"]["keywords"]
score = 0
factors: List[str] = []
med_count_val: Optional[float] = None
for f in facts:
kw_rules = rules.get(f.keyword)
if not kw_rules:
continue
if kw_rules["type"] == "range" and f.is_numeric:
pts, label = self._score_range_keyword(kw_rules, f.value)
score += pts
if f.keyword == "Medication Count":
med_count_val = f.value
if pts > 0:
factors.append(f"{f.keyword}={f.value} ({label}, +{pts})")
elif kw_rules["type"] == "categorical":
val = str(f.value).lower().strip()
pts = kw_rules["values"].get(val, 0)
score += pts
if pts > 0:
factors.append(f"{f.keyword}={val} (+{pts})")
# Derived polypharmacy: if med_count >= 5 and Polypharmacy not already scored
polypharmacy_scored = any("Polypharmacy" in f for f in factors)
if med_count_val is not None and med_count_val >= 5 and not polypharmacy_scored:
score += 3
factors.append(f"Derived Polypharmacy (Med Count={int(med_count_val)} >=5, +3)")
return ClusterScore("MEDICATIONS", min(score, 15), 15, factors)
def score_procedures(self, facts: List[ParsedFact]) -> ClusterScore:
rules = self._scoring_rules["PROCEDURES"]["keywords"]
score = 0
factors: List[str] = []
specific_scored = False
for f in facts:
kw_rules = rules.get(f.keyword)
if not kw_rules:
continue
if f.keyword == "Mechanical Ventilation":
# Mixed type: numeric > 0 or categorical
if f.is_numeric and f.value > 0:
score += kw_rules["score_if_any_positive"]
factors.append(f"Mechanical Ventilation={f.value} days (+{kw_rules['score_if_any_positive']})")
specific_scored = True
elif str(f.value).lower().strip() != "no":
score += kw_rules["score_if_any_positive"]
factors.append(f"Mechanical Ventilation={f.value} (+{kw_rules['score_if_any_positive']})")
specific_scored = True
elif f.keyword == "Dialysis":
val = str(f.value).lower().strip()
pts = kw_rules["values"].get(val, 0)
score += pts
if pts > 0:
factors.append(f"Dialysis={val} (+{pts})")
specific_scored = True
elif f.keyword == "Surgery":
val = str(f.value).lower().strip()
pts = kw_rules["values"].get(val, 0)
score += pts
if pts > 0:
factors.append(f"Surgery={val} (+{pts})")
specific_scored = True
elif f.keyword == "Any Procedure":
# Only score if no specific procedure was scored
pass # handled below
# Fallback: Any Procedure
if not specific_scored:
for f in facts:
if f.keyword == "Any Procedure":
val = str(f.value).lower().strip()
pts = rules["Any Procedure"]["values"].get(val, 0)
score += pts
if pts > 0:
factors.append(f"Any Procedure={val} (generic fallback, +{pts})")
break
return ClusterScore("PROCEDURES", min(score, 15), 15, factors)
def score_utilization(self, facts: List[ParsedFact]) -> ClusterScore:
rules = self._scoring_rules["UTILIZATION"]["keywords"]
score = 0
factors: List[str] = []
for f in facts:
if not f.is_numeric:
continue
kw_rules = rules.get(f.keyword)
if not kw_rules:
continue
pts, label = self._score_range_keyword(kw_rules, f.value)
score += pts
if pts > 0:
factors.append(f"{f.keyword}={f.value} ({label}, +{pts})")
return ClusterScore("UTILIZATION", min(score, 20), 20, factors)
def score_disposition(self, facts: List[ParsedFact]) -> ClusterScore:
rules = self._scoring_rules["DISPOSITION"]["keywords"]
score = 0
factors: List[str] = []
for f in facts:
kw_rules = rules.get(f.keyword)
if not kw_rules:
continue
val = str(f.value).strip()
if f.keyword == "Discharge Disposition":
val = self._normalize_discharge_disposition(val)
elif f.keyword == "Mental Status":
val = self._normalize_mental_status(val)
# Try exact match first, then case-insensitive
pts = kw_rules["values"].get(val, kw_rules["values"].get(val.lower(), 0))
score += pts
if pts > 0:
factors.append(f"{f.keyword}={val} (+{pts})")
return ClusterScore("DISPOSITION", min(score, 15), 15, factors)
# -- Layer 4: Pattern Detector ------------------------------------------
def detect_interactions(
self,
facts: Dict[str, List[ParsedFact]],
cluster_scores: Dict[str, ClusterScore],
) -> List[InteractionResult]:
"""Detect cross-cluster clinical patterns."""
results: List[InteractionResult] = []
# Helper: get numeric value for a cluster/keyword
def get_val(cluster: str, keyword: str) -> Optional[float]:
for f in facts.get(cluster, []):
if f.keyword == keyword and f.is_numeric:
return f.value
return None
def get_str(cluster: str, keyword: str) -> Optional[str]:
for f in facts.get(cluster, []):
if f.keyword == keyword:
return str(f.value).lower().strip()
return None
def has_symptom_group(group_id: str) -> bool:
for f in facts.get("SYMPTOMS", []):
val = str(f.value).lower().strip()
if val in ("yes", "severe"):
g = self.map_symptom_to_group(f.keyword)
if g and g["id"] == group_id:
return True
return False
def has_problem_group(group_id: str) -> bool:
for f in facts.get("PROBLEMS", []):
val = str(f.value).lower().strip()
if val in ("chronic", "acute", "exist"):
g = self.map_problem_to_group(f.keyword)
if g and g["id"] == group_id:
return True
return False
# --- Sepsis Pattern ---
hr = get_val("VITALS", "Heart Rate")
sbp = get_val("VITALS", "Systolic BP")
rr = get_val("VITALS", "Respiratory Rate")
wbc = get_val("LABS", "WBC")
temp = get_val("VITALS", "Temperature")
if hr is not None and hr > 100:
has_hemodynamic = (sbp is not None and sbp < 100) or (rr is not None and rr > 22)
has_infection = (
(wbc is not None and (wbc > 12 or wbc < 4))
or (temp is not None and temp > 100.4)
)
if has_hemodynamic and has_infection:
results.append(InteractionResult(
"sepsis_pattern", "Sepsis / SIRS Pattern", 10,
f"HR={hr}, SBP={sbp}, RR={rr}, WBC={wbc}, Temp={temp}",
))
# --- AKI Pattern ---
cr = get_val("LABS", "Creatinine")
bun = get_val("LABS", "BUN")
k = get_val("LABS", "Potassium")
na = get_val("LABS", "Sodium")
bicarb = get_val("LABS", "Bicarbonate")
if cr is not None and cr > 1.5 and bun is not None and bun > 30:
has_electrolyte = (
(k is not None and k > 5.0)
or (na is not None and na < 135)
or (bicarb is not None and bicarb < 22)
)
if has_electrolyte:
results.append(InteractionResult(
"aki_pattern", "Acute Kidney Injury Pattern", 8,
f"Cr={cr}, BUN={bun}, K={k}, Na={na}, Bicarb={bicarb}",
))
# --- Decompensated HF ---
if has_problem_group("heart_failure"):
has_decomp_sign = (
has_symptom_group("edema_fluid")
or has_symptom_group("respiratory_distress")
or (bun is not None and bun > 40)
)
if has_decomp_sign:
results.append(InteractionResult(
"decompensated_hf", "Decompensated Heart Failure", 8,
"Heart failure + fluid overload/dyspnea/elevated BUN",
))
# --- Frailty Syndrome ---
age = get_val("DEMOGRAPHICS", "Age")
hgb = get_val("LABS", "Hemoglobin")
mental = get_str("DISPOSITION", "Mental Status")
disp = get_str("DISPOSITION", "Discharge Disposition")
n_problem_groups = len(set(
self.map_problem_to_group(f.keyword)["id"]
for f in facts.get("PROBLEMS", [])
if str(f.value).lower().strip() in ("chronic", "acute", "exist")
and self.map_problem_to_group(f.keyword) is not None
))
if age is not None and age > 75:
frailty_count = 0
if n_problem_groups >= 3:
frailty_count += 1
if hgb is not None and hgb < 10:
frailty_count += 1
if mental in ("confused", "lethargic"):
frailty_count += 1
if disp in ("snf", "ltac", "rehab"):
frailty_count += 1
if frailty_count >= 2:
results.append(InteractionResult(
"frailty_syndrome", "Frailty Syndrome", 6,
f"Age={age}, problems={n_problem_groups}, Hgb={hgb}, mental={mental}, disp={disp}",
))
# --- Unstable Discharge ---
if disp == "ama":
results.append(InteractionResult(
"unstable_discharge", "Unstable Discharge (AMA)", 5,
"Discharge Against Medical Advice",
))
elif mental in ("confused", "lethargic") and disp in ("home", None):
results.append(InteractionResult(
"unstable_discharge", "Unstable Discharge (altered + Home)", 5,
f"Mental={mental}, Disposition={disp}",
))
# --- Respiratory Failure ---
spo2 = get_val("VITALS", "SpO2")
if spo2 is not None and spo2 < 92:
has_resp = (rr is not None and rr > 24) or has_symptom_group("respiratory_distress")
if has_resp:
results.append(InteractionResult(
"respiratory_failure", "Respiratory Failure Pattern", 6,
f"SpO2={spo2}, RR={rr}",
))
# --- Metabolic Crisis ---
glucose = get_val("LABS", "Glucose")
if glucose is not None and glucose > 300:
has_metabolic = (
(bicarb is not None and bicarb < 18)
or (k is not None and k > 5.5)
)
if has_metabolic:
results.append(InteractionResult(
"metabolic_crisis", "Metabolic Crisis (DKA/HHS)", 6,
f"Glucose={glucose}, Bicarb={bicarb}, K={k}",
))
# --- Bleeding Risk ---
plt = get_val("LABS", "Platelet")
anticoag = get_str("MEDICATIONS", "Anticoagulation")
if hgb is not None and hgb < 8:
has_bleed_risk = (
(plt is not None and plt < 100)
or anticoag == "yes"
)
if has_bleed_risk:
results.append(InteractionResult(
"bleeding_risk", "Active Bleeding Risk", 6,
f"Hgb={hgb}, Plt={plt}, Anticoag={anticoag}",
))
return results
# -- Layer 5: Risk Aggregator -------------------------------------------
def _logistic(self, score: int) -> float:
"""Convert composite score to probability via logistic function."""
z = self._alpha + self._beta * score
return 1.0 / (1.0 + math.exp(-z))
def _classify_risk(self, score: int) -> Tuple[str, str]:
"""Return (category, color) for a given composite score."""
for cat in self._scoring_rules["_meta"]["risk_categories"]:
if cat["score_min"] <= score <= cat["score_max"]:
return cat["name"], cat["color"]
return "Critical", "red"
# -- Layer 6: Days Predictor --------------------------------------------
def _predict_days(self, score: int) -> float:
"""Estimate days to readmission (point estimate)."""
return max(1.0, self._d_max * math.exp(-self._gamma * score))
def _predict_bucket(self, estimated_days: float) -> str:
if estimated_days <= 7:
return "0-7 days"
elif estimated_days <= 14:
return "8-14 days"
else:
return "15-30 days"
def _predict_survival(self, score: int, p_30d: float) -> SurvivalCurve:
"""Compute P(readmit by day t) for several horizons."""
k = self._k_base + 0.02 * (score - 30)
k = max(0.5, k) # floor to avoid degenerate cases
horizons: Dict[int, float] = {}
denom = 1.0 - math.exp(-k)
if abs(denom) < 1e-9:
denom = 1e-9
for t in [7, 14, 21, 30]:
f_t = (1.0 - math.exp(-(t / 30.0) * k)) / denom
p_t = p_30d * f_t
horizons[t] = round(min(max(p_t, 0.0), 1.0), 4)
return SurvivalCurve(horizons=horizons)
# -- Main Scoring Pipeline -----------------------------------------------
def score(self, facts: Dict[str, List[ParsedFact]], n_parsed: int = 0, n_dropped: int = 0) -> RiskResult:
"""Run full scoring pipeline on parsed facts."""
# Layer 3: Cluster scores
cluster_scores: Dict[str, ClusterScore] = {}
cluster_scores["DEMOGRAPHICS"] = self.score_demographics(facts.get("DEMOGRAPHICS", []))
cluster_scores["VITALS"] = self.score_vitals(facts.get("VITALS", []))
cluster_scores["LABS"] = self.score_labs(facts.get("LABS", []))
cluster_scores["PROBLEMS"] = self.score_problems(facts.get("PROBLEMS", []))
cluster_scores["SYMPTOMS"] = self.score_symptoms(facts.get("SYMPTOMS", []))
cluster_scores["MEDICATIONS"] = self.score_medications(facts.get("MEDICATIONS", []))
cluster_scores["PROCEDURES"] = self.score_procedures(facts.get("PROCEDURES", []))
cluster_scores["UTILIZATION"] = self.score_utilization(facts.get("UTILIZATION", []))
cluster_scores["DISPOSITION"] = self.score_disposition(facts.get("DISPOSITION", []))
# Layer 4: Interaction detection
interactions = self.detect_interactions(facts, cluster_scores)
interaction_bonus = sum(i.bonus for i in interactions)
# Layer 5: Aggregate
composite = sum(cs.score for cs in cluster_scores.values()) + interaction_bonus
probability = self._logistic(composite)
category, color = self._classify_risk(composite)
# Layer 6: Days prediction
est_days = self._predict_days(composite)
bucket = self._predict_bucket(est_days)
survival = self._predict_survival(composite, probability)
# Explainability
risk_factors: List[str] = []
protective_factors: List[str] = []
for cs in cluster_scores.values():
risk_factors.extend(cs.contributing_factors)
# Identify protective factors (normal values in important clusters)
for cluster in ["VITALS", "LABS"]:
cs = cluster_scores[cluster]
if cs.score == 0 and facts.get(cluster):
protective_factors.append(f"Normal {cluster.lower()} at discharge")
if cluster_scores["DISPOSITION"].score == 0 and facts.get("DISPOSITION"):
protective_factors.append("Stable disposition (Home, alert)")
for i in interactions:
risk_factors.append(f"[PATTERN] {i.pattern_name} (+{i.bonus})")
# Missing data
missing_clusters = [c for c in VALID_CLUSTERS if c not in facts or not facts[c]]
completeness = 1.0 - len(missing_clusters) / len(VALID_CLUSTERS)
if completeness >= 0.7:
confidence = "high"
elif completeness >= 0.5:
confidence = "medium"
else:
confidence = "low"
return RiskResult(
composite_score=composite,
cluster_scores=cluster_scores,
interaction_bonus=interaction_bonus,
interactions_triggered=interactions,
probability=round(probability, 4),
risk_category=category,
risk_color=color,
estimated_days=round(est_days, 1),
days_bucket=bucket,
survival_curve=survival,
risk_factors=risk_factors,
protective_factors=protective_factors,
missing_clusters=sorted(missing_clusters),
data_completeness=round(completeness, 2),
confidence=confidence,
n_facts_parsed=n_parsed,
n_facts_dropped=n_dropped,
)
# -- Convenience Methods ------------------------------------------------
def score_from_toon(self, toon_text: str) -> RiskResult:
"""Score from raw TOON text."""
facts, n_parsed, n_dropped = self.parse_toon(toon_text)
return self.score(facts, n_parsed, n_dropped)
def score_from_file(self, path: Union[str, Path]) -> RiskResult:
"""Score from a TOON text file."""
text = Path(path).read_text(encoding="utf-8")
return self.score_from_toon(text)
def score_from_jsonl(self, path: Union[str, Path], limit: int = 0) -> List[Tuple[str, RiskResult]]:
"""Score all entries in a JSONL file (trainset_full format).
Returns list of (hadm_id, RiskResult).
"""
results: List[Tuple[str, RiskResult]] = []
p = Path(path)
with p.open("r", encoding="utf-8") as f:
for i, line in enumerate(f):
if limit and i >= limit:
break
obj = json.loads(line)
hadm_id = str(obj.get("hadm_id", f"row_{i}"))
completion = obj.get("completion", "")
if completion:
result = self.score_from_toon(completion)
results.append((hadm_id, result))
return results
# ---------------------------------------------------------------------------
# Pretty-printing
# ---------------------------------------------------------------------------
def format_result(result: RiskResult, hadm_id: str = "") -> str:
"""Format RiskResult as human-readable report."""
lines: List[str] = []
header = f"=== Readmission Risk Report"
if hadm_id:
header += f" (hadm_id: {hadm_id})"
header += " ==="
lines.append(header)
lines.append("")
# Summary
lines.append(f"RISK: {result.risk_category} ({result.risk_color})")
lines.append(f"Probability of 30-day readmission: {result.probability:.1%}")
lines.append(f"Composite score: {result.composite_score}")
lines.append(f"Confidence: {result.confidence} (data completeness: {result.data_completeness:.0%})")
lines.append("")
# Days prediction
lines.append("--- Days-to-Readmission Prediction ---")
lines.append(f"Point estimate: ~{result.estimated_days:.0f} days")
lines.append(f"Bucket: {result.days_bucket}")
lines.append("Survival curve:")
for t, p in sorted(result.survival_curve.horizons.items()):
lines.append(f" P(readmit by day {t:2d}): {p:.1%}")
lines.append("")
# Cluster breakdown
lines.append("--- Cluster Scores ---")
for cluster in ["DEMOGRAPHICS", "VITALS", "LABS", "PROBLEMS", "SYMPTOMS",
"MEDICATIONS", "PROCEDURES", "UTILIZATION", "DISPOSITION"]:
cs = result.cluster_scores.get(cluster)
if cs:
lines.append(f" {cluster}: {cs.score}/{cs.max_score}")
lines.append(f" INTERACTIONS: +{result.interaction_bonus}")
lines.append(f" TOTAL: {result.composite_score}")
lines.append("")
# Risk factors
if result.risk_factors:
lines.append("--- Risk Factors ---")
for rf in result.risk_factors:
lines.append(f" - {rf}")
lines.append("")
# Protective factors
if result.protective_factors:
lines.append("--- Protective Factors ---")
for pf in result.protective_factors:
lines.append(f" + {pf}")
lines.append("")
# Triggered patterns
if result.interactions_triggered:
lines.append("--- Clinical Patterns Detected ---")
for ix in result.interactions_triggered:
lines.append(f" [{ix.pattern_id}] {ix.pattern_name}: +{ix.bonus} pts")
lines.append(f" Evidence: {ix.description}")
lines.append("")
# Missing data
if result.missing_clusters:
lines.append(f"--- Missing Data ({len(result.missing_clusters)} clusters) ---")
for mc in result.missing_clusters:
lines.append(f" ? {mc}")
lines.append("")
lines.append(f"Facts parsed: {result.n_facts_parsed}, dropped: {result.n_facts_dropped}")
return "\n".join(lines)
# ---------------------------------------------------------------------------
# CLI
# ---------------------------------------------------------------------------
def main():
import argparse
ap = argparse.ArgumentParser(description="Rule-based 30-day readmission risk engine")
sub = ap.add_subparsers(dest="cmd")
# Score a single TOON file
p_file = sub.add_parser("file", help="Score a single TOON file")
p_file.add_argument("path", help="Path to TOON text file")
# Score from JSONL
p_jsonl = sub.add_parser("jsonl", help="Score all entries in a JSONL file")
p_jsonl.add_argument("path", help="Path to JSONL file")
p_jsonl.add_argument("--limit", type=int, default=0, help="Limit number of entries")
p_jsonl.add_argument("--summary", action="store_true", help="Show summary statistics only")
# Score from inline TOON text
p_inline = sub.add_parser("inline", help="Score inline TOON text (pipe to stdin)")
args = ap.parse_args()
engine = ReadmissionRiskEngine()
if args.cmd == "file":
result = engine.score_from_file(args.path)
print(format_result(result))
elif args.cmd == "jsonl":
results = engine.score_from_jsonl(args.path, limit=args.limit)
if args.summary:
scores = [r.composite_score for _, r in results]
probs = [r.probability for _, r in results]
categories = {}
for _, r in results:
categories[r.risk_category] = categories.get(r.risk_category, 0) + 1
print(f"=== Summary ({len(results)} patients) ===")
print(f"Score: mean={sum(scores)/len(scores):.1f}, "
f"min={min(scores)}, max={max(scores)}, "
f"median={sorted(scores)[len(scores)//2]}")
print(f"P(readmit): mean={sum(probs)/len(probs):.1%}")
print("Risk categories:")
for cat in ["Low", "Medium", "High", "Critical"]:
n = categories.get(cat, 0)
pct = n / len(results) * 100 if results else 0
print(f" {cat}: {n} ({pct:.0f}%)")
days = [r.estimated_days for _, r in results]
print(f"Days estimate: mean={sum(days)/len(days):.1f}, "
f"min={min(days):.1f}, max={max(days):.1f}")
else:
for hadm_id, result in results:
print(format_result(result, hadm_id))
print("\n" + "=" * 60 + "\n")
elif args.cmd == "inline":
import sys
toon_text = sys.stdin.read()
result = engine.score_from_toon(toon_text)
print(format_result(result))
else:
ap.print_help()
if __name__ == "__main__":
main()
|