Potiential-Automation/src/cms1500_text.py

"""
CMS-1500 (02/12) reader that uses the PDF's own text layer.

Waystar's ViewClaimPDF returns digital PDFs: every value is real text at a
fixed spot on the form. Reading those words by position is exact and runs
offline, so no OCR and no PHI leaves the machine.

All coordinates are normalized to 0-1 (x from the left, y from the top) and
taken from Waystar's CMS1500_0212 rendering.
"""
import re
from decimal import Decimal
from dataclasses import dataclass


@dataclass(frozen=True)
class Word:
    text: str
    x: float  # horizontal mid-point, 0-1
    y: float  # vertical mid-point, 0-1


# ================= BOX 21 — DIAGNOSIS CODES =================
_DX_SLOT_LABEL = re.compile(r"^([A-L])\.$")
_ICD10 = re.compile(r"^[A-Z]\d[0-9A-Z]{1,5}$")
_DX_Y_MIN, _DX_Y_MAX = 0.590, 0.640
_DX_X_MAX = 0.615      # box 22/23 starts to the right
_DX_ROW_TOL = 0.010    # codes print a little above their slot label


def parse_diagnosis_codes(words):
    """Return [(slot_letter, code), ...] ordered A..L."""
    in_box = [w for w in words if _DX_Y_MIN <= w.y <= _DX_Y_MAX and w.x <= _DX_X_MAX]
    labels = [(m.group(1), w) for w in in_box if (m := _DX_SLOT_LABEL.match(w.text))]
    found = {}
    for w in in_box:
        if not _ICD10.match(w.text):
            continue
        same_row = [(s, lw) for s, lw in labels
                    if abs(lw.y - w.y) <= _DX_ROW_TOL and lw.x <= w.x]
        if not same_row:
            continue
        slot, _ = max(same_row, key=lambda sl: sl[1].x)  # nearest label to the left
        found.setdefault(slot, w.text)
    return sorted(found.items())


# ================= BOX 24 — SERVICE LINES =================
@dataclass(frozen=True)
class ServiceLine:
    line_no: int
    dos_from: str
    dos_to: str
    pos: str
    cpt: str
    modifiers: tuple
    dx_pointer: str
    charge: object  # Decimal, or None when not printed
    units: str
    npi: str


_CPT = re.compile(r"^(\d{5}|[A-V]\d{4}|\d{4}[TF])$")
_LINES_Y_MIN, _LINES_Y_MAX = 0.665, 0.848   # below the box-24 header, above box 25
_LINE_ROW_TOL = 0.006                       # shaded supplemental row sits ~0.009 above

# Column x-ranges (on the word's mid-point), left to right.
_COLS = {
    "from_mm": (0.035, 0.066), "from_dd": (0.066, 0.101), "from_yy": (0.101, 0.137),
    "to_mm":   (0.137, 0.171), "to_dd":   (0.171, 0.205), "to_yy":   (0.205, 0.240),
    "pos":     (0.240, 0.280),
    "cpt":     (0.315, 0.405),
    "mods":    (0.405, 0.545),
    "pointer": (0.545, 0.615),
    "charge":  (0.615, 0.712),
    "units":   (0.712, 0.760),
    "npi":     (0.825, 0.960),
}


def _in_col(row, col):
    lo, hi = _COLS[col]
    return [w.text for w in sorted(row, key=lambda w: w.x) if lo <= w.x < hi]


def _date(mm, dd, yy):
    if not (mm and dd and yy):
        return ""
    year = yy if len(yy) == 4 else f"20{yy}"
    return f"{mm.zfill(2)}/{dd.zfill(2)}/{year}"


def parse_charge(text):
    """'9000' -> 90.00 (Waystar prints dollars and cents without a point)."""
    text = text.replace("$", "").replace(",", "").strip()
    if not re.fullmatch(r"\d+(\.\d{1,2})?", text):
        return None
    return Decimal(text).quantize(Decimal("0.01")) if "." in text else Decimal(text) / 100


def parse_service_lines(words):
    """Return ServiceLine for each box-24 row that has a CPT/HCPCS code, top to bottom."""
    section = [w for w in words if _LINES_Y_MIN <= w.y <= _LINES_Y_MAX]
    lo, hi = _COLS["cpt"]
    anchors = sorted((w for w in section if lo <= w.x < hi and _CPT.match(w.text)),
                     key=lambda w: w.y)
    lines = []
    for anchor in anchors:
        row = [w for w in section if abs(w.y - anchor.y) <= _LINE_ROW_TOL]
        first = lambda col: (_in_col(row, col) or [""])[0]
        charge_text = "".join(_in_col(row, "charge"))
        lines.append(ServiceLine(
            line_no=len(lines) + 1,
            dos_from=_date(first("from_mm"), first("from_dd"), first("from_yy")),
            dos_to=_date(first("to_mm"), first("to_dd"), first("to_yy")),
            pos=first("pos"),
            cpt=anchor.text,
            modifiers=tuple(_in_col(row, "mods")),
            dx_pointer="".join(_in_col(row, "pointer")),
            charge=parse_charge(charge_text) if charge_text else None,
            units="".join(_in_col(row, "units")),
            npi=first("npi"),
        ))
    return lines


# ================= HEADER BOXES (1a, 3, 26, 28) =================
_BOX = {  # (x_min, x_max, y_min, y_max)
    "insured_id":      (0.600, 0.990, 0.135, 0.155),
    "dob":             (0.380, 0.500, 0.166, 0.182),
    "patient_account": (0.300, 0.470, 0.862, 0.876),
    "total_charge":    (0.615, 0.745, 0.862, 0.876),
}


def _box_words(words, box):
    x0, x1, y0, y1 = _BOX[box]
    return [w.text for w in sorted(words, key=lambda w: (round(w.y, 3), w.x))
            if x0 <= w.x <= x1 and y0 <= w.y <= y1]


def parse_header(words):
    dob = [t for t in _box_words(words, "dob") if t.isdigit()]
    total = "".join(t for t in _box_words(words, "total_charge") if t != "$")
    return {
        "insured_id": "".join(_box_words(words, "insured_id")).upper(),
        "dob": _date(*dob) if len(dob) == 3 else "",
        "patient_account": " ".join(_box_words(words, "patient_account")),
        "total_charge": parse_charge(total) if total else None,
    }


# ================= WHOLE CLAIM =================
@dataclass
class Claim:
    insured_id: str
    dob: str
    patient_account: str
    total_charge: object  # Decimal or None
    dx: list              # [(slot_letter, code), ...]
    lines: list           # [ServiceLine, ...]
    issues: list          # human-readable review notes; empty = clean


def build_claim(pages):
    """Build one Claim from the word lists of every page of a claim PDF."""
    if not any(pages):
        return Claim("", "", "", None, [], [], ["No text layer on PDF (scanned?) - check by hand"])

    header = parse_header(pages[0])
    lines, totals = [], []
    for words in pages:
        for ln in parse_service_lines(words):
            lines.append(ServiceLine(**{**ln.__dict__, "line_no": len(lines) + 1}))
        totals.append(parse_header(words)["total_charge"])
    total = None if None in totals else sum(totals, Decimal("0.00"))
    claim = Claim(header["insured_id"], header["dob"], header["patient_account"],
                  total, parse_diagnosis_codes(pages[0]), lines, [])
    claim.issues = _review(claim)
    return claim


def _review(claim):
    issues = []
    if not claim.insured_id:
        issues.append("No Insured ID (box 1a)")
    if not claim.dob:
        issues.append("No Date of Birth (box 3)")
    if not claim.dx:
        issues.append("No diagnosis codes (box 21)")
    if not claim.lines:
        issues.append("No service lines found")
    slots = {s for s, _ in claim.dx}
    for ln in claim.lines:
        for label, value in (("Date of Service", ln.dos_from), ("POS", ln.pos),
                             ("Units", ln.units), ("Charge", ln.charge),
                             ("Rendering NPI", ln.npi)):
            if not value:
                issues.append(f"Line {ln.line_no}: no {label}")
        for letter in ln.dx_pointer:
            if letter not in slots:
                issues.append(f"Line {ln.line_no}: Dx pointer {letter} has no diagnosis code")
    line_sum = sum((ln.charge or Decimal("0.00") for ln in claim.lines), Decimal("0.00"))
    if claim.total_charge is None:
        issues.append("No total charge (box 28)")
    elif line_sum != claim.total_charge:
        issues.append(f"Line charges {line_sum:.2f} != box 28 total {claim.total_charge:.2f}")
    return issues


# ================= REPORT ROW =================
def _per_line(values, label):
    if len(values) == 1:
        return values[0]
    return "; ".join(f"{label} {i}: {v}" for i, v in enumerate(values, 1))


def to_report_row(claim, claim_id="", insurance="", patient_name="",
                  signature_physician="", fallback_account="", fallback_charges=""):
    """One report row per claim: the bot's original 15 columns + Line Charges + Review Notes."""
    lines = claim.lines
    money = lambda d: f"{d:.2f}" if d is not None else ""
    dx_codes = [code for _, code in claim.dx]
    return {
        "Claim ID":               claim_id,
        "Rendering Provider #":   _per_line([l.npi for l in lines], "RP #"),
        "Signature of Physician": signature_physician,
        "Patient's Account No":   claim.patient_account or fallback_account,
        "Date of Birth":          claim.dob,
        "Insurance Name":         insurance,
        "Insured ID Number":      claim.insured_id,
        "Patient's Name":         patient_name,
        "Date of Service":        _per_line([l.dos_from for l in lines], "DOS"),
        "Diagnosis Codes":        f"{', '.join(dx_codes)} (Total: {len(dx_codes)})" if dx_codes else "",
        "CPT":                    _per_line([l.cpt for l in lines], "CPT"),
        "POS":                    _per_line([l.pos for l in lines], "POS"),
        "Modifier":               _per_line([" ".join(l.modifiers) for l in lines], "Modifier"),
        "Units":                  _per_line([l.units for l in lines], "Units"),
        "Charges":                money(claim.total_charge) or fallback_charges,
        "Line Charges":           _per_line([money(l.charge) for l in lines], "Charge"),
        "Review Notes":           "; ".join(claim.issues),
    }


# ================= PDF ENTRY POINT =================
def read_pdf_pages(pdf_path):
    """Word lists (normalized positions) for every page of the PDF."""
    import pdfplumber
    pages = []
    with pdfplumber.open(str(pdf_path)) as pdf:
        for page in pdf.pages:
            width, height = float(page.width), float(page.height)
            pages.append([
                Word(w["text"], (w["x0"] + w["x1"]) / 2 / width, (w["top"] + w["bottom"]) / 2 / height)
                for w in page.extract_words()
            ])
    return pages


def extract_claim(pdf_path):
    return build_claim(read_pdf_pages(pdf_path))