cs.CLOct 13, 2026
SaveFrom Qualitative Text to Integrated Evidence: The Parallel-Then-Integrated (PTI) Procedure
Organizations: Department of Emergency Medicine, Yale School of Medicine, Yale University, New Haven, CT, USA · Department of Social and Behavioral Sciences, Yale School of Public Health, Yale University, New Haven, CT, USA
Abstract
Mixed methods research offers established strategies for integrating qualitative and quantitative evidence, but less guidance for integrating two computational representations of the same qualitative corpus. This methodological study introduces Parallel-Then-Integrated (PTI), a four-step procedure for explicit and traceable integration. We present PTI using 1,203 interview paragraphs from 29 mothers in Nepal. The two representations were ontology-guided Patient Voice (PV) coding with PVminer and data-driven themes generated through non-negative matrix factorization (NMF). A theme-complete crosswalk documented convergence, divergence, and preprocessing-sensitive differences. Depression status, based on Patient Health Questionnaire-9 (PHQ-9) scores, served as a secondary external anchor for exploratory association checks. PTI provides a structured approach for integrating computational representations of qualitative text within mixed methods research.
Figures & tables
| Check | What it examines | Illustrative evidence |
| Domain uniqueness | Domains represented by only one representation | Representation-specific domains |
| Feature overlap | Association between features from the two representations | Cross-representation associations |
| Signal increment | Added signal when an external anchor is available | Comparison with the individual representations |
| Interpretive richness | Insights produced through integrated interpretation | Meta-inferences drawing on both representations |
Table 1: Heuristic checks for examining relationships between PTI representations
| Stage | Mothers | Rows | Removed | Per-mother counts | ||
| Min | Mean | Max | ||||
| 29-mother analytic cohort | 29 | 4,253 | — | 83 | 146.7 | 233 |
| Participant turns only | 29 | 1,242 | 3,011 | 12 | 42.8 | 95 |
| After short-turn filter ( 5 words) | 29 | 1,203 | 39 | 12 | 41.5 | 93 |
Table 2: Preprocessing funnel: transcript rows to analytic paragraphs
Figure 1: Overview of the PTI demonstration. Participant-response paragraphs are transformed into two independent computational representations (PVminer and NMF), then aligned, assessed for overlap, and integrated through the joint display. Cross-validated logistic regression provides a secondary external-anchor association check. The joint display is the primary analytic output.
| PV domain | Content focus/sub-codes |
| Care coordination | Patient/provider coordination of care |
| Partnership | Active participation/involvement, express opinions, state preferences, alignment, appreciation/gratitude, connection, salutation, build trust |
| Shared decision-making | Explore options, seeking approval, approval of decision, share options, make decision, approval/reinforcement |
| SDOH | Economic stability, education access and quality, health care access and quality, neighborhood and built environment, social and community context |
| Socioemotional behavior | Express emotions/concern/unease, sadness/fear, acknowledge situation, reassurance, positive remarks/reward |
Table 3: Patient Voice ontology domains
| Theme | Label | Content focus | Top TF–IDF terms |
| T0 | Baby care and health | Infant care, monitoring illness, and providing care | baby, care, care baby, look, look baby, sick, birth, provide |
| T1 | Emotional difficulty and help-seeking | Feeling difficulty and asking for or receiving help | feel, difficult, feel difficult, didn, didn feel, help, don feel, ask |
| T2 | Activities and preferences | Leisure, play, likes, and dislikes | like, things, play, don like, talk, activities like, works, activities |
| T3 | Family and in-laws | Husband, parents, sisters, and extended family | mother, in-law, mother in-law, sister, husband, sister in-law, father, father mother |
| T4 | Denial and negation | Denying problems, uncertainty, and negation talk | don, problems, don problems, umm, think, know, don know, things |
| T5 | Daily chores and hygiene | Washing, clothing, food, and household time use | time, clothes, used, wash, wash clothes, washing, food, washing clothes |
Table 4: Participant-response NMF themes ( )
| Representation | Features | Evaluation | AUC | Permutation |
| PV model (standalone) | 5 | LOOCV | 0.706 | 0.106 |
| NMF themes (standalone) | 6 | LOOCV | 0.680 | 0.090 |
| Combined PV + NMF | 5 PV + up to 4 NMF | LOOCV | 0.756 | 0.050 |
Table 5: LOOCV AUC by representation ( mothers)
| NMF theme | Mapped PV domain(s) | Alignment | Joint-display decision |
| T0: Baby care and health | Care coordination; SDOH / SocialAnd CommunityContext | Partial convergence; no dedicated infant-care construct | Excluded: already covered by existing care and social-context codes. |
| T1: Emotional difficulty and help-seeking | Socioemotional behavior | Partial convergence | Included as Emotional difficulty ; NMF adds a difficulty-oriented context not specified by the PV construct. |
| T2: Activities and preferences | Partnership/active participation and involvement | Partial convergence | Excluded from joint display: shared activity/engagement focus, but PV is care/interaction-oriented while T2 is everyday likes, play, and preferences; routine disruption maps to T5. |
| T3: Family and in-laws | Partnership, SDOH / EconomicStability | Convergence with Partnership, partial convergence with SDOH | Included as two joint-display rows: Economic hardship and Partnership and family . |
| T4: Denial and negation | None | Apparent divergence under default preprocessing | Excluded: failed contraction-normalization sensitivity; retained as a methodological finding. |
| T5: Daily chores and hygiene | None | Divergence: NMF-only | Included as Routine and daily-life disruption . |
Table 6: Theme-complete cross-stream crosswalk: NMF themes mapped to relevant PV domains, with joint-display inclusion decisions
| Domain | PV code–sub-code | NMF | Alignment | Meta-inference |
| Economic hardship | SDOH / EconomicStability | T3 | Partial convergence: NMF captures hardship through family and husband language. | Economic hardship is voiced relationally, not as isolated financial talk. |
| Partnership and family | PartnershipPatient / expressOpinions | T3 | Convergence: both representations capture relational content. | In the illustrated account, distress is voiced through conflict with the husband. |
| Emotional difficulty | SocioEmotional Behaviour | T1 | Partial convergence: both representations capture affective content, while NMF adds a difficulty-oriented context. | Distress is expressed through felt difficulty and help-seeking. |
| Routine and daily-life disruption | None | T5 | Divergence: NMF captures routine-disruption language absent from the ontology. | Routine-related language may be incompletely represented by the ontology, indicating a candidate coverage gap requiring validation. |
Table 7: Joint display of cross-stream alignment and meta-inferences (PTI Steps 2–4)