Person-centred, integrated and pro-active care for multi-morbid elderly with advanced care needs: a propensity score-matched controlled trial

Even though international recommendations highlight the complex and multi-faceted synergies between these three PIP-elements, studies in this field have generally followed the traditional approaches of the “rational-linear” approach, “… When designing intervention and implementation strategies, as well as when conducting rigorous evaluations, there is a tendency to reduce messy real world situations into the individual component parts in an attempt to determine the relationships between them” [20, 26]. However, to achieve better outcomes for persons with CLNs, the whole chain of care needs to change. We propose that the PIP-components are inter-dependent, and will only produce the expected effects if all the components are in place. The effect of an intervention where the PIP elements work synergistically together is mostly unexplored.

PCC is in many ways the foundation of PIP-care. PCC is widely recognised as desirable [13, 15, 27] yet the gap between the rhetoric and reality of PCC remains uncomfortably wide [28, 29]. The last few years, a sharper vision of PCC described by Bisognano as “flipped healthcare” is emerging [15, 30, 31]. It flips care focus from: What is the matter?" to “What matters to you?” It forces attention from professionally defined diagnosis and functional goals to personal goals. Here, we prefer the term “Goal-oriented Person-centred care” (Goal-PCC), to underline that in “flipped care,” the entire chain of care, including care planning, delivery, and evaluation is co-created with the person and driven by personal goals [32]. While Coulter found in her Cochrane review that Goal-PCC had a small but significant impact on outcomes in single disease contexts, neither Coulter nor we have found any studies that applied this concept to multi-morbid patients [15].

Norway has a robust primary care system [33], with high accessibility and low out of pocket expenses, among the highest life expectancies in the world, and excellent health outcomes in international comparisons, and a commitment to quality improvement for persons with CLNs [34‐39]. However, persons with CLNs in Norway report the same challenges as other high-income countries, making the Norwegian context internationally relevant [39‐41].

We designed the PACT intervention to address quality challenges identified by patients with CLNs in Norway [41], in alignment with the theoretical and empirical foundations of chronic care ideals. Central ideals were the Chronic Care Model [42], Goal-oriented Person-centred care [15, 31, 41, 43, 44], Shared Decision Making [45], integrated care pathways [46] and pro-active risk management [23, 47]. We were mindful of the complex nature of both the health service and the PACT intervention [48‐50]. The care process for each person, here denoted as the individualised Patient Pathway (iPP) [41], is a product of a complex adaptive system (CAS) [49]. Key elements to support problem-solving in a CAS are: A good enough vision or goal, simple guiding rules and a wide space for innovation and creativity to reach common goals [49]. The goals and rules were set out in our previous work on the Person-centred integrated care quality framework [41]. Briefly, the iPP serves patient-defined goals [31, 44], with attention to PCC, IC and Pro-C, while incorporating insights and feedback from both the patient and other contributors is an ongoing attribute of the care process [48]. To the best of our knowledge, we have not found other studies on the effects of a complex, synergistic and “flipped” Goal-PCC model, which builds all three PIP-elements into the care model the way our Patient-Centred Team (PACT) intervention does.

The PACT study aims to improve the triple aim of patient care experience, health outcomes and cost-benefit ratios for persons with CLNs. Due to delays in the collection of patient reported outcomes (see details later), we here report on early and available health outcomes using routine data from the electronic health record (EHR). We expect that the PACT intervention applied to a population of high risk frail multi-morbid elderly will stabilize the patient’s health situation through improved quality of care. As a result, we expect to see a reduced use of high-level emergency care and a higher use of low-level planned care. Hypothetically, use of emergency care could be reduced simply by denying patients adequate care. This would of course be unethical, and in this frail group, we would expect increased mortality. We therefore needed to show that reductions in the use of high-level emergency care was not accompanied by increased mortality. To summarize, the hypotheses (H) examined in this paper are that the PACT intervention will cause:

This study is a prospective propensity score (PS) matched controlled trial evaluating the effectiveness of the PACT service. Briefly, the use of routine data from the EHR, and the propensity score techniques described by Rubin [51‐53], allowed us to create a valid comparison group even when the target population was defined by a discretionary referral process. The same EHR data also contained relevant process and outcome measures, which enabled a comparison of intervention and usual care. We follow recommendations for evaluation of complex interventions [54], complexity science [55] and quality improvement methodology [56].

The study design, which has been called a synthetic RCT, lies somewhere between the randomized controlled trial (RCT) and the observational study. We adhere to the CONSORT [57] STROBE [58], TIDieR [59] and the SAMPL [60] checklists for reporting on RCT, observational, intervention and biostatistical studies respectively.

We have previously published our study protocol at ClinicalTrials.gov (identifier: NCT02541474) and as a protocol paper [61]. We found it necessary to make the following protocol changes: 1) Due to delays in data collection, we report on secondary health outcomes before we are able to report on the primary outcome: the patient reported health experience by the SF12 questionnaire [62]. 2) Due to a high risk of healthy selection bias, we were allowed to include routine EHR data on all patients receiving PACT without their consent, while the original protocol planned inclusion based on written informed consent.

Patients, who had consented to receive care (care consent) from the PACT team, were eligible to be invited to give their informed written consent to participate in the PACT study (study consent). However, a high proportion suffered from acute serious or overwhelming morbidity at baseline. It is unethical to ask patients to fill a study consent or questionnaire in a situation where treatment and adjustment should take precedence. When patients had recovered sufficiently for study recruitment to be ethical, many were already past the 4 week study recruitment window and were no longer eligible. 2) A high proportion of persons with cognitive impairment were unable to provide informed consent and we found that family members were frequently uncomfortable answering personal questions on the patient’s behalf. For these two reasons, the recruitment and collection of the primary outcome based on patient reported data is delayed, is still ongoing and will continue to March 2019.

As early results on secondary health outcomes were important for health service decision makers regarding the funding of the PACT service, we looked for other ways to evaluate the PACT service. We also feared that a study based on informed consent would be flawed by a healthy selection bias. With this background, we sought and were granted permission (see ethics section) to extract routinely collected data from the EHR which contains both exposure and outcomes variables for both the intervention and control population, without asking for patient consent. This would ensure fair recruitment, irrespective of challenges to provide consent. The main ethical concern of this approach was to maintain privacy of the routine data provided to the research team. The privacy protection officer approved the project with the following precautions to ensure the privacy of the participants: We prepared pseudonymised analysis files, which means that direct identifiers such as names, addresses or id-numbers, were replaced with a pseudonym. We saved and used analysis files exclusively on “offline” devices/ computers. Only 3 researchers (GB, JH, TB1) had access to the analysis files. All intermediate and final analysis results are fully anonymous.

We, are therefore able to report here on the outcomes available to us from the EHR: healthcare utilisation and risk of death.

The intervention was tailored to each patient, as long as the PACT team adhered to the following structures and principles.

The study extracted following variables for eligible populations in the four intervention and control municipalities from the EHR: Age, sex, start and stop time of outpatient, ambulatory care and inpatient episodes of care, degree of emergency for each episode, ICD-10 diagnoses and Diagnosis Related Group (DRG)-codes registered for each episode [64], DRG-points and date of death. Professionals collect these variables as a matter of routine care for all patients. The local hospital, which supplied our EHR-data, provides 97% or more of all out-patient and in-patient hospital services for the local population [65]. All Norwegian hospitals report these data routinely to the Norwegian Patient Registry (NPR). NPR consistently assesses data quality and corrects errors in the source data in collaboration with the source institution [65]. The EHR is updated with mortality data from the Norwegian Population Registry (PR) on a monthly basis, including all deaths (i.e. deaths at home, hospital or other institutions) in the study population. It takes approximately one week from when the physician issues a death certificate until the PR is updated (personal communication from PR).

We followed intention to treat principle, as we included all patients who were offered and gave their consent to receive services from PACT services in analyses, independent of the actual length, content, and delivery of PACT services.

530 persons were referred to the PACT-intervention for 606 care episodes from Oct. 2014 – Sept. 2016. The PACT-referral review team declined 5 % of referrals. Figure 1 shows the exclusion flowchart. 83% of all referred and 94% of all eligible persons contributed to final analyses.

The dataset showed an excellent balance, with no balance tests outside of the recommended range. Some of the matching variables differed significantly between groups (see Table 2). As expected, the distribution of Lead days, which we were unable to match for, differs between groups.

All adjusted RR indicated a reduction of high-level emergency care utilisation.

In our strictest sub-group analysis, RR for emergency admissions was 0.90 (95%CI: 0.82–0.99), RR for sum emergency bed days was 0.68 (95%CI: 0.52–0.79) and RR for 30-days readmissions 0.72 (95%CI: 0.41–1.24). See Table 3 for details of multivariate analyses in other groups.

The adjusted RR indicated a substantial increase in planned low-level care and no change in low-level emergency care.

The RR for planned outpatient visits was 2.3 (95%CI: 2.02–2.55), while RR for emergency outpatient visits was 0.9 (95%CI: 0.68–1.20) in our strictest sub-group analysis. Please see Table 3 for details of multivariate analyses in other groups.

The adjusted RR indicated a reduction in mortality risk at both three and six months follow-up.

The RR for death at three months was 0.39 (95% CI: 0.22–0.70) in our strictest sub-group analysis. The protection waned somewhat after three months so that at six months the RR was 0.57 (95% CI: 0.34–0.94), also in our strictest sub-group analysis. Please see Table 3 for details of multivariate analyses for other groups. The Kaplan–Meier plot shows a net survival benefit in the intervention groups by the end of the follow-up period (Fig. 2).

There were in all 28 referred patients who did not get the PACT intervention within 24 h and were therefore ineligible for analyses. Of these 6 (21%) died. There were in all 4 (15%) deaths among the 26 persons who did receive the intervention but were excluded from analyses due to technical reasons (no prior hospital history, no suitable match). Small numbers preclude drawing any conclusion from these numbers.

Compared with propensity score matched controls, the care process of frail multi-morbid elderly who received the PACT intervention had a reduced risk of high-level emergency care, increased use of low-level planned care, and substantially reduced mortality risk.

Several comparable interventions in the literature pro-actively identify frail patients in GP EHR, but these studies fail to show effects [95, 96]. It may be, that it is not enough to capture frailty, as some very frail persons may be balancing their lives well enough, and interventions may upset their precarious balance. Finding formal methods of capturing the frail person as they move to an unstable state, while there is still time to regain balance, is an important research question. A formal definition of the PACT target population would also allow us to evaluate how outcomes for a formally identified frail elderly in intervention municipalities compared to the same group in control municipalities.

To further examine external validity in an international context, a comparison of the Norwegian results with international propensity score matched control-groups is an aim for future research.

Although we attribute effects to our theoretical underpinnings, the study merits further evaluation of the causes of death in the two treatment groups, to understand better how effects are produced [97]. A stronger understanding of the care process and mechanisms of effect and how patient pathways differ between groups and contexts would help us understand which components require high fidelity in any intervention site, and which components can and should tailor to the local context.