Qualitative analysis of the vocabulary used in work logs of a preventive programme for elderly oral function and nutrition

Abstract Background In Japan, day care services for elders include programmes aimed at improving nutrition and oral and motor functions. Few studies have qualitatively assessed these interventions. Objective To qualitatively search for the characteristic words used in the work logs of a preventive programme on oral function and nutrition for elders by intervention period and intervention type． Methods We included 83 participants (81.3 ± 8.2 years) from four day care services in Japan and divided them into the following groups randomly: those who received oral function intervention only, nutritional intervention only and those who received combined oral function plus nutritional intervention. The interventions were conducted twice per month for 24 months. Data from handwritten work logs were entered into a computer as text files. Monitoring of frequently appearing words, co‐occurrence analysis and cross‐tabulation by intervention period and intervention types was conducted using text mining analysis. Results Correspondence analysis revealed that the words used during 1‐6 months and 7‐12 months were similar in participants’ subjective content, and those used in objective content in 13‐18 months and 19‐24 months were similar. These results indicate that subjective improvements increased after 13 months, and it was maintained within 24 months. The combined intervention type is ideal for oral and nutrition problems. Conclusion Because this text mining approach revealed the changes in the words used and could be used to monitor any subjective improvement, this approach may help evaluate the effects of preventive care.


| BACKG ROU N D
The average lifespan of Japanese people has continued to increase; in 2013, the average lifespans of men and women were 80.2 and 86.6 years, respectively. However, the average healthy lifespans of men and women are 71.2 and 74.2 years, respectively. 1 Thus, overall average lifespans and healthy lifespans differ by approximately 10 years. To extend healthy lifespans, it is critical to prevent individuals from declining to a state in which they require long-term care. The effects of these programmes are as follows: improvements in swallowing function (assessed using repetitive saliva swallowing test (RSST)), oral diadochokinesis 2 and lip closure force. 3 In addition, Morishita et al 4 reported that a combined intervention programme significantly improved vitality index and pronunciation of the syllable/pa/ in oral diadochokinesis. However, there may be some benefits to preventive care, which are difficult to assess using objective numbers, such as facial expressions, attitudes and emotions.
The text mining approach has been used with various types of qualitative data, such as written records by nurses during disasters, 5 interview data related to multidisciplinary cooperation among medical professionals 6 and free-response content in class assessment surveys. 7 If the text mining approach is applied to the records of the preventive care programme, other effects, which are difficult to evaluate by an objective numerical value, may become obvious. In addition, by investigating the frequency of occurrence of words, it may be possible to obtain new insights into these programmes. However, no attempt has been made to apply text mining analysis to preventive care programmes. Therefore, we aimed to analyse preventive care service work logs of the intervention study conducted by Morishita et al 4 to identify the characteristic words used by intervention period and intervention type.

| Programmes for improving oral function and nutrition
In four elderly day care services in Aichi Prefecture, Japan, 130 elderly were agreed to participate in this study. A total of 35 elderly were excluded because of severe dementia (clinical dementia rating score was 3), poor health condition and hospitalisation as described by Morishita et al 4 We enrolled 95 participants in our study (35 men and 60 women; mean age 82.7 ± 6.9 years). The participants were randomly divided into three groups using a computer-generated sequence until the numbers of the participants were equal in all groups: those who received oral function intervention only, those who received nutritional intervention only and those who received a combination of both oral function and nutritional intervention. The interventions for each group were implemented from November 2012 to October 2014. At the end of the study, 83 participants remained. The reasons why the other participants were excluded from the study were as follows: hospitalisation, admission to a nursing home and death. For the 31 participants (15 men and 16 women) in the oral function-only group, seven dental hygienists provided oral function programme; instructions on oral hygiene, exercise for expression muscles, tongue, salivary glands, swallowing and dysarthria training. For the 23 participants (nine men and 14 women) in the nutrition-only group, five registered dietitians provided nutrition programme; self-check of appetite and dehydration, diet balance and weight management. For the last 29 participants (nine men and 20 women) in the combined oral function and nutrition group, both the dental hygienists and the registered dietitians implemented their respective programmes. The same dental hygienists or registered dietitians were in charge of both the single programme and the combined programme. To unify the programme contents, the dental hygienists and registered dietitians received training. The interventions were conducted twice a month over a 24-month period.

| Text mining approach
Text mining techniques were used to analyse the work logs recorded by the seven dental hygienists and five registered dietitians as they conducted their programmes. Text mining is a content analysis method that uses quantitative analysis techniques to sort or analyse qualitative text data. 8 The use of quantitative techniques to analyse qualitative data prevents arbitrary interpretations on the part of the analyst. The work logs were handwritten by the dental hygienists or registered dietitians just after each programme session for recording and for messages to other performers. It was made for each participant. One log per session consisted of 200-300 words. They were instructed to describe the records, which were divided into four categories: subjective contents, objective contents, assessments and programme contents. Subjective contents included words used by the elderly individuals themselves, and objective contents included the objective findings of the dental hygienists or registered dietitians. In this study, only subjective and objective contents were used in the analysis.
Except for personal data that would allow identification of individual participants (like names and addresses), the data in all handwritten work logs were entered into the computer as a text file and were then analysed using KH Coder (available from http://khc.sourc eforge.net/). 9 We also used ChaSen analysis software for preprocessing and to perform morphological analysis. Compound words such as "full dentures," "remaining teeth," "oral cavity" and "Patakara" (a lip trainer) were adopted as force pickup words. At first, frequencies of words were estimated in subjective contents and objective contents. Next, a correspondence analysis was performed to observe the characteristic words by intervention period (1-6 months, 7-12 months, 13-18 months and 19-24 months) and intervention type (oral function-only, nutrition-only and combined). Then, categorisation was performed depending on the type of cluster analysis.
Finally, cross-tabulation was performed by intervention period and intervention type, using the categories. A P-value of <5% was considered significant.

| E THIC AL CONS IDER ATIONS
This study was conducted with the approval of the Institutional Review Board of the National Center for Geriatrics and Gerontology (application no. 605). Potential participants received information on the purpose and content of the study before the start of the study; those who provided consent were enrolled as participants.
Additionally, all staff who took part in this study provided consent.
All data were anonymised and analysed under the condition that individual participants could not be identified.

| Characteristic words and co-occurrences
In total, 375 238 words (31 189 sentences) were obtained from the handwritten data, with the exclusion of personal information. By categories of records, 138 329 words (20 824 sentences) were used as subjective contents, whereas 115 269 words (20 706 sentences) were used as objective contents.
In the subjective contents, the number of different words, excluding the pure appearance words without duplication, was 6430.
The most frequently used word was eat occurring 1320 times, followed by say (590 times) and meal (490 times). The remaining 10 most characteristic words were go (428 times), drink (358 times), night (352 times), denture (346 times), home (312 times), good (296 times) and tongue (289 times). The most co-occurring word was morning, followed by brushing, eat and meal.
In the objective contents, the number of different words, excluding the pure appearance words without duplication, was 5868. The most frequently used word was tongue coating (901 times), followed by denture (695 times Note: The value listed on the right side of each word represents the Jaccard index. A high Jaccard index indicates strong co-occurrence relation. Table 1 shows the results of the analysis of characteristic words used during the intervention types. The value listed on the right side of each word represents the Jaccard index, rated from 0 to 1. A high Jaccard index indicates a strong characteristic of the word by cooccurrence analysis. In the subjective contents, the words related to meal and eat were considered as strong characteristic words and indicated that the elders cared about meals. In 1-6 months and 7-12 months, denture was on a centred position. After 7 months, the centre words that the evaluator was cared about were tongue coating and plaques in the objective contents.

| Analysis by intervention periods
In the correspondence analysis, the appearance patterns of the words in the subjective contents were divided into three phases: The words used at 13-18 months and 19-24 months demonstrated an almost similar appearance pattern. Of the words, the co-occurrences for tongue coating and plaque were particularly strong.
"Oral function" had a good appearance rate in 13-18 months (9.5%), than in 1-6 months (7.1%), 7-12 months (8.6%) and 19-24 months (8.9%). The correspondence analysis of the subjective contents showed that the characteristic words in the combined intervention appeared at the centre of the oral function-only intervention and nutrition-only intervention. The same tendency was observed for objective contents.

| Analysis by intervention type
In the subjective contents, the words demonstrating co-occurrence in all intervention types were say, meal and eat. However, in the objective contents, none of the words had a co-occurrence relation to all three interventions, although some words had co-occurrence relationship between oral function-only and combined intervention or between nutrition-only and combined intervention. The words eat, say, meal and lunch had strong co-occurrences between nutrition-only and combined interventions, whereas the words tongue coating, denture, plaque and clean revealed strong co-occurrences between oral function-only and combined interventions.
The results of the cross-tabulation analysis of the subjective contents using categories shown in Table 2 are indicated in Figure 1B.
The appearance frequencies of all categories were significantly different depending on the intervention type. "Oral hygiene" was more frequently observed in the oral function-only intervention (18.2%) than in the nutrition-only (6.4%) and combined intervention (13.1%).

| D ISCUSS I ON
This is the first study to assess the work logs of preventive care services using text mining approach. The words such as "daily life," "meal" and "oral hygiene," which were categorised using cluster analysis, were frequently used throughout the 24-month period. In other words, elders had cared about "daily life," "meal" and "oral hygiene" within the two-year intervention period. On the basis of the correspondence analysis of the objective contents, the focus of the staff was moved from "oral hygiene" in 7-12 months to "oral function" in 13-18 months. In the early stages, the staff focused first on "oral hygiene." After improvements were observed, they focused on "oral function." The correspondence analysis also revealed that the F I G U R E 1 Appearance rate of categories by intervention period (A, C) and intervention type (B, D). Subjective contents were shown in A and B, objective contents were shown in C and D. The larger the square, the higher the appearance rate in the whole. The shading of colour indicates at which intervention period (or type), the appearance rate is high among the same category. The number in y-axis indicated the appearance times. * P < 0.01; χ2 test.  words used in 1-6 months and 7-12 months were similar in subjective contents. In 13-18 months and 19-24 months, the words that indicated a good condition such as fine and fun were observed. In addition, the words that appeared in 13-18 months and 19-24 months were similar in objective contents. These results may indicate that subjective improvements had increased after 13 months and hence another set of interventions may be started and maintained within 24 months. In this study, we divided the intervention period into six months based on a previous study, which had reported improvements in bite force, RSST and salivary flow rate after a six-month preventive care programme. 10 However, no study has reported on the suitable intervention period for such analyses. To monitor the changes over time, the data should be collected and analysed on a monthly basis.
Analysis of characteristic words revealed that the most frequently used word was eat, followed by say and meal in the subjective contents, and tongue coating, denture and tongue in the objective contents. There were eight types of words that were related to dentures: full dentures, full false teeth, dentures, partial dentures, partial false teeth, false teeth, FD and PD. In a study of comprehension of dentistry terms conducted among care workers, 100% of the participants understood the term dentures, whereas only 80% of the participants understood the terms full dentures and partial dentures. 11 The use of technical terms will not become a problem if the preventive care programme is conducted only by one discipline. However, when the care becomes multidisciplinary, replacing technical terms with lay expressions may be important for proper sharing of information.
A correspondence analysis by intervention type showed that the characteristic words in the combined intervention appeared at the centre of the oral function-only intervention and nutritiononly intervention. A cross-tabulation analysis revealed that the words used in the oral function-only or nutrition-only intervention were biased; however, the words were used in a well-balanced manner in the combined intervention. When evaluated using objective numbers, a combined intervention programme significantly improved vitality index and pronunciation of the syllable/pa/ in oral diadochokinesis. 4 In addition, in the study conducted in 2010, we combined a programme for improving motor function and improving nutrition with a programme for improving oral function.
Consequently, compared with unidisciplinary programmes, a higher percentage of participants in the combined oral function plus nutrition programme had a lower need for nursing care, as well as a reduced risk of health decline to a level requiring care (ie, reduced risk of falls, fractures, aspiration pneumonia and the like), suggesting that the combined programme yielded a greater preventive care effect. 12 Therefore, there is a possibility that the combination intervention type is ideal to cover oral and nutrition problems.
This study had some limitations. The number of participants who could not finish the 24-month continued intervention was higher in the nutrition-only intervention type. Furthermore, the bias in the number of participants led to the bias in the number of records and thus influenced the results of the analysis. Another limitation was that the staff involved in the single and combined intervention type overlapped. In the future, the analysis should be conducted on the basis of the difference between occupation type and intervention form. To make the difference in occupation clear, the single intervention and combined intervention should be performed by different staffs without concurrent operation.

| CON CLUS ION
The text mining approach was applied to the work logs of preventive care services. It was suggested that the elders cared about "daily life," "meal" and "oral hygiene." A correspondence analysis revealed that subjective improvements had increased after 13 months and hence another set of interventions must be performed and this condition must be maintained within 24 months. In addition, there was a possibility that the combination intervention type was ideal to cover oral and nutrition problems. Because this text mining approach revealed changes in the word used and showed subjective improvements, it may contribute to the evaluation of the effect of preventive care.

ACK N OWLED G M ENTS
We are grateful for the collaboration of the Nishikasugai