Mild Cognitive Impairment, Reversion Rates, and Associated Factors: Comparison of Two Diagnostic Approaches

Study population

Data were drawn from the Swedish longitudinal population study “Good Aging in Skåne” (GÅS), including participants ranging in age from 60–102 years [31]. GÅS is part of the national study “The Swedish National Study on Aging and Care” (SNAC) and has been ongoing since 2001. Participants from five municipalities in southern Sweden are randomly selected from the Swedish population register, and a new baseline cohort is added every sixth year. Participation rates varies between 60–70%. Follow-up visits were made every third year for those older than 78 years and every sixth year for those younger than 78. Participant data from three baseline waves, and their associated follow-up examinations, were pooled for the purpose of this study.

Between 2001 and 2020, 5,804 participants were examined, and during this period 1,785 and 987 were classified as MCI according the Petersen and Neuropsychological criteria, respectively. This classification event was considered their MCI baseline visit, regardless of which study wave it occurred in. As the NP criteria insists on two impaired scores per domain, fewer participants could be included in the NP sample due to incomplete cognitive data. To inspect rates and factors for reversion, these 1,785/987 participants were followed until they either: 1) reverted, 2) no longer met MCI criteria (ADL worsened, no longer reported a cognitive complaint), 3) progressed to dementia, 4) were lost to follow-up, whichever event occurred first (see Figs. 1 and 2 for further clarification). For the analysis inspecting dementia progression, reverters, and stable MCI cases were followed until they received a dementia diagnosis from the Skåne Healthcare register, which was obtained until 31 December 2020. The study was approved by the regional Ethics Committee of Lund University, and all participants signed a written informed consent.

Fig. 1

A flow chart of participant selection, drop-out, reversion of mild cognitive impairment when applying the Petersen criteria. Participants were selected from three baseline waves and were followed until they met MCI criteria, either in their first visit or follow-up examinations. Once they received their MCI diagnosis they were then followed until they were lost to follow-up, were not eligible for MCI at any of the follow-ups (i.e., their ADL worsened, progressed to dementia etc.), continued to be MCI or reverted. ADL, activities of daily living; MCI, mild cognitive impairment; CDR, clinical dementia rating scale

Fig. 2

A flow chart of participant selection, drop-out, reversion of mild cognitive impairment when applying the Neuropsychological criteria. Participants were selected from three baseline waves and were followed until they met MCI criteria, either in their first visit or follow-up examinations. Once they received their MCI diagnosis they were then followed until they were lost to follow-up, were not eligible for MCI at any of the follow-ups (i.e., their ADL worsened, progressed to dementia etc.), continued to be MCI or reverted. ADL, activities of daily living; MCI, mild cognitive impairment; CDR, clinical dementia rating scale

Using standardized procedures, participants filled in self-rated questionnaires, performed cognitive tests, and were examined by a physician and a nurse. An experienced test administrator conducted neuropsychological tests assessing four cognitive domains: memory (a 16-item word-recall test, a 16-word recognition test, and a 5-object delayed recall test), speed of processing (a digit cancellation test and a pattern comparison test), verbal fluency (a word fluency test of animals and professions, respectively), and executive function (a modified version of the trail making test B and the digit span backwardstest) [32].

Diagnostic criteria of MCI and reversion

An algorithmic approach, blind to prior diagnosis, was applied retrospectively. In this study, both MCI definitions have the same four requirements: a self- or informant reported cognitive complaint, objective cognitive impairment in one or more cognitive domains, normal or minimally impaired daily function, and no dementia. Researchers applying the NP criteria usually omit the requirement of the cognitive complaint [5, 15, 16], as it has been associated with elevated misclassification rates [4]. Nevertheless, as informant or subjective based complaint is a core requirement in both the original MCI definition, and in the definition mild neurocognitive disorder (Diagnostic and Statistical Manual of Mental Disorders 5th edition) and is a common reason for referral to memory clinics, this requirement is retained for both criteria. The NP criteria applied here is consequently considered a modified version. In this study, the two definitions differ in the number of required impaired test results within a cognitive domain, and their cut-off score. The Petersen criteria requires at least one impaired test score, within a cognitive domain, and below the 7th percentile relative to a normative sample. In contrast, the NP criteria requires a score below the 16th percentile, on at least two tests, within the same cognitive domain. To measure multiple domain MCI, both criteria required impaired test scores (≥1 for Petersen and ≥2 for NP within the same domain) on at least two cognitive domains. Normative values were derived from two sister studies, SNAC-Blekinge and SNAC-Stockholm, which administrated the same cognitive tests in a standardized way (Supplementary Table 1) [33]. A test score below the 7th/16th percentile, corrected for age, sex and education, was considered an impaired test score. When assuming distribution normality, the 7th/16th percentile equates to 1.5/1.0 SD below the mean.

Subjective cognitive complaint was assessed by inquiring if the participant experienced that their memory had worsened, and the informed complaint was determined by the examining physician. To assess the daily function of the participants both basic activities of daily living (bADL) and instrumental ADL (iADL) were considered. bADL and iADL were determined using the Katz index of independence [34], and a modified Lawton-Brody index [35], respectively. For MCI classification bADL was preserved, while impairment in one area was allowed for iADL [2]. The dementia diagnosis was classified by the study physician according to the Diagnostic and Statistical Manual of Mental Disorders 4th edition. These diagnoses were complemented with information from the participants’ medical records and diagnosis codes reported in the Skåne Healthcare Register, including Alzheimer’s disease, mixed dementia, vascular dementia, Lewy body dementia, and frontotemporal dementia.

Reverters were defined as participants that fulfilled all criteria for MCI, but no longer met the requirement for objective cognitive impairment, i.e., the participants had normal ADL, cognitive complaint, and no dementia, but their cognitive scores no longer fell within the range of impairment. Stable MCI was defined as participants who continued to be MCI throughout their study participation. Reverters and stable MCI were defined by both criteria separately. To calculate reversion rate, the proportion of reverters was compared to the proportion of non-reverters, i.e., the stable MCI group.

Factors for reversion

The following factors for reversion were considered: age, sex, physical activity, cohabitation, sleeping disturbances, alcohol consumption, body mass index (BMI), smoking status, total years of formal education, depressive symptoms, cardio- and cerebrovascular diseases, hypertension, diabetes mellitus type I and II, and number of impaired cognitive domains, all measured at the time of the first MCI classification.

Physical activity, cohabitation, smoking status, alcohol consumption, and years of formal education were self-reported. Participants were asked how much (during the last year) they moved and activated their bodies during their spare time and in their home, and their answers were defined accordingly: sedentary (sedentary or being mostly still) or being physically active (light to moderate exercise or heavy exercise). Cohabitation was defined as living with someone (unrelated to, if it was a spouse, partner, or child, or someone else). Smoking was defined as either current or previous smoker, or never smoked. Participants were asked how often they consumed alcohol, and their answers were categorized accordingly: no/low consumption (never or once a month, or less frequently), moderate consumption (2–4 times per month), or high consumption (2-3 times per week or more frequently).

Sleep disturbances were reported to the physician and included one or more conditions: sleep latency, waking up multiple times during the night, early awakenings, sleep apnea, or other bothersome sleeping issues.

Comorbidities were identified by several methods: self-reported to the study physician, through medical examination, and by reviewing medical records. In case of a discrepancy between the different sources, information recorded in the study database was used and in case of a missing value the information recorded in the healthcare registry was implemented. Diagnosis codes classified according to the International Classification of Diseases system version 10 were retrieved from the Skåne county healthcare registry. Cardiovascular disease was defined as: at least once being diagnosed with acute myocardial infarction, heart failure, ischemic heart disease, angina pectoris, presence of cardiac and vascular implants, heart failure, cardiac arrhythmias, cerebral infarction, nontraumatic intracranial hemorrhaged, occlusion and stenosis of precerebral or cerebral arteries, transient cerebral ischemia, and/or hypertension. Depression was measured through interview by the test administrator, using the Montgomery Åsberg Depression Rating Scale [36].

Statistical analyses

Comparison of dementia risk between reverters and stable MCI

We investigated how MCI reversion affected the likelihood of developing dementia, using a Cox proportional hazard regression model, with the occurrence of the first reversion as the only explanatory variable, which was modelled as a time-varying covariate. The median time to reversion, and the duration of follow-up, were also calculated using a Cox regression model.

For all analyses, the statistical significance level was set to 5%. The statistical analyses were performed using Stata SE 17 software (StataCorp LLC, College Station, TX, USA) and Python 3.8.5 (Python Software Foundation).

MCI diagnosis and follow-up

Of the 5,804 participants that were included in the study, 1,785/987 (Petersen criteria/NP criteria) were considered MCI in at least one study visit. Of those participants a further 1,258/663 had at least one follow-up visit with 744/375 eligible to fulfil the applied MCI criteria at a subsequent follow-up visit (see flow chart 1 and 2 for further clarification of eligibility). The median time for follow-up for Petersen criteria was 7.0 (95% CI: 6.8; 7.6) years and 6.7 (95% CI: 6.5; 6.9) years for NP criteria). Seen irrespective of applied MCI criteria, a majority of MCI participants had one impaired cognitive domain (≥70%) in a non-amnestic cognitive domain (≈60%). An impaired cognitive domain was at least one (Petersen criteria) or two (NP criteria) cognitive test scores below the appointed cut-off score in a single cognitive domain. Mean age for the first MCI diagnosis was 68 years of age, and women were slightly overrepresented regardless of the applied criteria (Table 1).

Reversion rates in respective criteria

Of the 744 MCI participants, according to the Petersen criteria, 323 (43.4%, 95% CI: 39.8; 47.0) reverted and the incidence rate of reversion was estimated to 63.0 (95% CI: 57.4; 69.1) person/1000 year. The NP criteria produced similar reversion rates, with 181 of 375 (48.3%, 95% CI: 43.2; 53.5) reverting, and the incidence rate of reversion was estimated to 76.2 (95% CI: 67.9; 85.6) person/1000 year. The results of the sensitivity analyses were comparable with those observed in the primary analyses, with reversion rates varying between 42.1–45.3%. However, excluding cognitive complaint, as required in the original NP criteria, produced somewhat higher reversion rates (54.7%). See Table 2 for specifics. Calculation of reversion rates for the Petersen and the NP criteria, for a sample containing the same participants (n = 334), yielded a slightly lower reversion rate of the Petersen criteria (32.9%), but similar for the NP criteria (46.4%) as in the primary analyses (Supplementary Table 4).

For the Petersen criteria, 111 of 742¹ with baseline MCI were later diagnosed with dementia. Reverters had a lower hazard ratio (HR) (0.34, 95% CI: 0.21; 0.54) of being diagnosed with dementia compared to those with stable MCI. For the NP criteria, 74 of 374 participants with baseline MCI were later diagnosed with dementia. Reverters had a lower HR (0.17, 95% CI: 0.1; 0.34) of being diagnosed with dementia compared to those with stable MCI (Fig. 3).

Fig. 3

Cumulative incidence of dementia for participants who developed MCI according to the Petersen and Neuropsychological criteria, respectively. The curves were calculated using a Cox proportional hazard model with the occurrence of the first reversion as the only explanatory variable, which was modelled as a time-varying covariate. Of note, there are two fewer participants with baseline MCI in this analysis compared to reversion analysis (2 for Petersen criteria and 1 for NP criteria) as they were classified as dementia according to medical records but not by study physician. In this case, medical records were used for dementia diagnosis.

Baseline characteristics for reverters and stable MCI

Baseline demographics and general characteristics are presented in Table 1. On inspection slight differences were observed between reverters and non-reverters. The reverter group contained more participants that were cohabitant, had lower blood pressure, were male, and consumed more alcohol (seen for both criteria). Noticeably, there were more smokers, and they were slightly younger in the reverter group when classified according to the NP criteria.

There were observed performance differences in the cognitive tests between the reverters and the stable MCI. The majority of participants in the reverter group had only one cognitive domain affected in comparison to the stable MCI group who had additional domains affected (Petersen criteria: 86.1% versus 58.4%, NP criteria: 84.5% versus 71.6%). Cognitive test scores were overall worse in the stable MCI group in comparison to the reverter group (Table 3).

Factors for reversion

Two separate multivariate Poisson regression models including all covariates were run for the Petersen and the NP criteria, respectively (Table 4). For the Petersen criteria, the models indicated that the number of affected cognitive domains, alcohol consumption, and age at classification, affected the incidence rate of reversion (Table 3). The probability of reversion for participants who had one impaired cognitive domain was 0.54 (95% CI: 0.50; 0.60). In contrast to participants with two or more impaired domains, the probability of reversion was lower 0.20 (95% CI: 0.15; 0.26). For non-consumers of alcohol, the probability of reversion was 0.35 (95% CI: 0.29; 0.41) compared to peers who reported moderate and high alcohol consumption, 0.51 (95% CI: 0.45; 0.58) and 0.49 (95% CI: 0.38; 0.59), respectively. The calculated probability of reversion for a 60-year-old participant was 0.38 (95% CI: 0.33; 0.42) while for a 70-year-old participant it was 0.46 (95% CI: 0.42; 0.50).

For the NP criteria, the model indicated that the number of affected cognitive domains, cohabitation, and BMI affect the incidence rate ratio of reversion. The probability of reversion for participants who had one impaired cognitive domain was 0.53 (95% CI: 0.47; 0.60). In contrast to participants with two or more impaired domains, the estimated probability of reversion was lower 0.32 (95% CI: 0.21; 0.42). For those living alone, the probability of reversion was 0.38 (95% CI: 0.30; 0.46) compared to peers cohabiting 0.54 (95% CI: 0.47; 0.62), respectively. The probability of reversion for a participant with a BMI of 18.5 kg/m², 24.9 kg/m², and 29.9 kg/m² was 0.62 (95% CI: 0.46; 0.79), 0.51 (95% CI: 0.45; 0.57), and 0.43 (95% CI: 0.37; 0.50), respectively.

Reversion rates

Our reversion rates of 42.4% and 48.3% are higher than the 25% established in a meta-analysis [10], although they are within the reported range of 4%–59%. Moreover, they are lower than our previously established rate [8], and similar to rates from two studies applying the Petersen criteria: 37% [22] and 38% [37]. The handful of studies examining reversion rates, when applying the NP criteria, found fairly low rates of 7.4%, 11.4%, 15.8%, and 19.2% [14–16, 38]. Three of these studies also made a direct comparison to the Petersen criteria, all of which produced higher reversion rates, arguing that the Petersen criteria is overinclusive, and that the NP criteria is better in detecting true MCI. Noticeably, the reverter comparison group (i.e., the non reverters) contained other trajectory groups than stable MCI (in all of the above studies). For example, Wong and colleagues [15] reported reversion rates of 7.4% for NP and 13.5% for Petersen, but the reverter comparison group contained progressors, those with stable MCI, and those with stable intact cognition. Remarkedly, taking the same number of reverters and only comparing them to stable MCI, rates appear different with a reversion of 30.7.% for the NP and 27.8% for the Petersen criteria, thereby questioning their conclusion that the Petersen criteria is overinclusive. Additionally, it is common practice to omit cognitive complaint as a requirement for a NP MCI classification. Our sensitivity analysis showed a higher reversion rate for the NP criteria when cognitive complaint was omitted (54.7% versus 48.3%). Although, the rate discrepancy is fairly minor, it questions the value in excluding this criterion, which we argue is important as it provides an important context when interpreting the impaired test result. Additionally, both our operationalizations were successful in identifying dementia progressors, as our stable MCI groups had a higher risk of dementia in both criteria compared to the reverter groups.

These inferences, together with our results, has implications. Firstly, the NP criteria does not necessarily, in comparison to Petersen, decrease the risk of false positives [4]. Secondly, the requirement of two tests for each cognitive domain may be superfluous, which is encouraging as extensive testing could, due to fatigue, lead to unjust impaired scores. Also, previous work using data from the GÅS study showed that less than two impaired test scores per cognitive domain was not associated with reversion [8]. Thirdly, additional features for the criterion objective cognitive impairment in the MCI definition could be of equal importance in determining stability and progression. For instance, test characteristics (e.g., the reliability and validity of instruments), total number of measured domains, or level of cut-off. Finally, when comparing reversion rates, researchers must be mindful of the comparison group in addition to how reversion is defined, as rates can vary substantially thereof.

The reversion rate for NP criteria was calculated using a smaller sample than for the Petersen criteria, as it required more cognitive data for MCI classification. The sensitivity analysis, using the same participants, showed that the Petersen criteria produced a reversion rate of 13.5 percentage points lower than for the NP criteria. This supports our previous inference that the Petersen criteria is not overinclusive in comparison to the NP criteria. The reversion definition applied here could be considered strict, as it required all the MCI criteria, except for objective cognitive impairment, to be fulfilled. This potentially leads to the exclusion of stable MCI and reverter cases at follow-up, since functional and self-experienced cognitive deterioration is part of normal aging [39]. Therefore, sensitivity analyses were performed with more lenient eligibility criteria for MCI reversion. The results showed similar reversion rates across reversion definitions, indicating the robustness of our primary results. Nevertheless, we urge researchers to provide detailed information on the inclusion criteria for the cognitively intact, stable MCI, and reverters, to enable accurate rate comparisons across studies.

Factors for reversion

We found that the factors associated with reversion were criteria dependent, and the only common feature was having fewer than two impaired cognitive domains. Our results support the evidence that more than one affected cognitive domain is of importance for the future MCI trajectory, as it likely reflects both a more advanced and severe stage of the cognitive impairment [40, 41]. Studies consistently find that multiple domain MCI is associated with dementia progression [39, 42, 43], as singular domain MCI is equally associated with reversion [8, 16, 24, 44]. In addition, multiple domain is more diagnostically stable, where fewer individuals go from multiple domain to singular domain than vice versa [7, 8]. In summation, our results imply that the number of domains affected may be more crucial for diagnostic stability, and further progression, than the number of impaired scores in each domain.

Similar to our findings, that cohabitation appears beneficial for reversion, being married has been found to increase the likelihood of reversion [23], whereas living alone is associated with MCI progression and accelerated onset [45]. We assume that individuals who live with someone, or are married, have a day-to-day social contact, and social interaction evidently decreases risk of dementia [28]. Mental and social stimulation accompanying cohabitation is argued to promote cognitive reserve [46], but also perhaps shield from further cognitive deterioration [23], as people living with someone are more likely to seek help or comply with treatment. Noticeably, one study found that unmarried individuals reverted more often in comparison to progressors [25]. Others have argued that cohabiting or being married will hinder detection of cognitive impairment due to compensating factors of the partner or family member, or even due to denial of their partner’s cognitive deterioration [47]. Suggestively, further research on the role of marital status and cohabitation on MCI stability is warranted.

In this study alcohol consumption, both moderate and high consumption, increased the likelihood of MCI reversion in comparison to no or little consumption, which is consistent with reports from previous reversion studies [24, 48]. Partially contradictory to this, other studies conclude the alcohol-cognitive decline relationship to be dose response dependent, where moderate intake decreases risk of MCI progression and heavy drinking elevates risk [49, 50]. Then it is unexpected that our non/light-consumer group had a lower chance of reverting in comparison to our heavy consumers. Two explanations can be offered. Firstly, the light/non consumers contained unhealthy participants with deteriorated cognition leading to reduced or complete refrainment of alcohol consumption due to health-related issues. Noticeably, no differences in risk of reversion for low and non-consumers was detected in a sensitivity analysis (Supplementary Material) indicating that these two groups were similar. Secondly, heavy consumers or abusers could have stopped drinking leading to better cognitive health and reversion [48]. Markedly, alcohol consumption could directly reflect a rich social life, which we have noted above to be beneficial for cognitive health in older age.

Lower BMI was significantly associated with reversion. Prior studies produce contradictory outcomes, showing that both higher and lower BMI is linked to dementia progression [51–53]. Although, previous studies have failed to link BMI specifically with reversion. In our study, the incidence rate of reversion was only slightly affected by BMI (IRR: 0.97) in a fully adjusted model concluding then that the relevancy of BMI warrants further examination.

Opposing most previous work (e.g., [8, 21, 25]), but not all [54], older age was associated with reversion. This probably reflects a study design feature where healthy individuals are more prone to attend follow-up examinations and therefore cognitively preserved older adults are overrepresented in our sample.

Although sleep disturbances are more common in those with MCI, in comparison to those with normal cognition [30], our study showed similar frequencies in reported sleeping disturbances between reverters and those with stable MCI. A proposed explanation involves the nature of the sleep disturbance, as impaired cognition is mostly seen with abnormal sleep duration (short or long sleep), sleep apnea, and sleep-fragmentation [55]. We used an aggregated sleep variable containing multiple sleep disturbances, possibly being too general to detect variability in sleeping patterns for reverters and non-reverters. Consequently, further research on this potential factor for reversion is warranted.

It is well established that higher educational attainment reduces dementia risk [28] and incident MCI. Yet many previous studies (e.g., [39, 56]), including ours, fail to find that higher level of education predicts reversion [8, 21, 24, 54]. Given the similarity in duration of formal education between reverters and stable MCI, and the high level of education (11.4 years), we were not able to conclude the educational effect on reversion. Moreover, by applying education-corrected norm-scoring, when classifying the impaired from the non-impaired group, differences are reduced.

Strengths and limitations

There are several strengths to this study. Firstly, the GÅS study enables research of several large cohorts that are followed longitudinally over a long period of time, which in turn generates a large sample size of MCI participants. Secondly, the data gathering and testing of the participants is consistent between visits, ensuring cognitive impairment to be assessed equally each time. Thirdly, participants medical history was complemented with registry data. As registering a patient’s diagnosis is mandatory by law, all participants that received a dementia diagnosis via healthcare, up until the end of registry follow-up, could be incorporated in the analysis. Notably, underdiagnosis of dementia is a common issue in elderly populations, and consequently some undetected dementia cases in this study were considered non-demented. Lastly, data from GÅSs’ sister studies enabled a robust normative group.

This study is not without limitations. About 30% were lost to follow-up. Frail participants are less likely to attend follow-up visits, which could produce higher reversion rates since the healthiest returned to follow-up visits. To mitigate this effect, home visits were offered. Moreover, main baseline characteristics between attenders and those lost to follow-up were contrasted (Supplementary Material). Participants lost to follow-up, and ineligible for the application of the chosen MCI criteria (missing data or ADL-dependent), were somewhat older but otherwise had similar characteristics as those remaining in the study. Therefore, attrition bias may only play a minimal role on reversion rates. A second limitation involves reversion due to practice effects. Our participants were followed until a reversion event occurred, exposing them to multiple testing sessions, potentially leading to practice effects that masked true decline [57]. Since we implemented an extensive test battery with a larger number of non-amnestic tests, the likelihood of finding non-amnestic MCI subjects is artificially high.

Furthermore, other explanations of cognitive impairment are overlooked when applying an algorithmic approach in comparison to a consensus approach, such as, e.g., stress, distraction during testing, or side-effects of medication. Although, it is not feasible to retrospectively apply a consensus approach to a sample of over 5,000 individuals. Lastly, the present study does not examine biomarkers, genetics, or brain imaging, which could strengthen MCI diagnosis.

Future research should explore if inverse risk factors for incident MCI, or progression, are the same for MCI reversion, and examine the roll of variation in predictors between examinations (e.g., reversal of depression, or better sleeping habits). Additionally, research could focus on the status and characteristics of individuals with MCI that are typically excluded from further analyses, as they no longer meet certain criteria for MCI, reversion, or dementia at follow-up. In our sample, this was a substantial number of MCI cases (35.9% and 38.3%, for the Petersen and the NP criteria, respectively). Recent work suggests that some of these individuals (e.g., those who did not fulfil the MCI criteria because their daily function declined) are actually characteristically similar to dementia progressors [39]. Thus, this unrepresentative group of individuals can provide valuable information on MCI trajectories.

Conclusion

This work found that the 42–48% of participants that met the MCI criteria reverted at some point after diagnosis. Our results suggest that two impaired tests within the same cognitive domain and with a lenient cut-off, were not necessarily superior in diagnostic stability than when applying one test and a less lenient cut-off. Nevertheless, the search of a gold standard in the operationalization of the MCI definition, especially in the context of number of tests and optimal cut-off scores, should continue. This is especially vital as poor diagnostic stability renders high reversion rates, poor identification of factors associated with reversion and dementia progression. One ultimate goal in MCI research is to achieve precise predictions for progression, in order to enable successful drug interventions [58]. Finally, our results express important aspects of daily social stimulation, the severity of the cognitive impairment, the role of alcohol consumption, BMI, and age on predicting the future path of those diagnosed with MCI.